TL;DR

Purposive Accelerationism: A Third Way

¶Most founders, executives and the politicians trying to keep up with them are offered a binary: you are either for acceleration and human flourishing, or you are a doomer. For many of us the choice does not sit easily, because it is drawn on one simple axis, faster or slower, where every check looks like a brake. A truer map of the terrain has three: yes speed, but also who holds power in the system and what governance mechanisms we have to keep them in balance. The corner that stands empty today is fast, yet governed and plural, and this book is written to fill it.

A Genealogy and Formal Critique of Accelerationism

¶Effective accelerationism has a poetic appeal, yet its contradictions leave an unease in those of us who ask what future all this effort and toil is steering towards. That led me on a year-long journey through its genealogy, which revealed the internal tensions to be a feature not a bug: the doctrine is a theology, a vibe and a meme, not a science of progress, and it ultimately amounts to a fatalism, a nihilism and, worse, a permission for anyone to do as they will, because technology and capitalism, fused into a techno-capitalism, are the thermodynamic will of the universe running itself down. Its boldest move is to deny any governance beyond the discipline of the market as heresy, in doing so contravening one of the foundational laws of cybernetics: a system without a governor, the mechanism that holds it to a set value, does not run free. The rival camp, a safety movement grown out of effective altruism, keeps the creed that the machine is accelerating and differs only on whose hand should hold it, so its remedy is not to stop the engine but to seat the state at the centre, a pause only to make the handover smooth. However an engine without a governor directly at the machine substrate runs away; not towards freedom but towards totalisation, by the market and the oligarch, by the state or, by a hybrid Leviathan where the two fuse.

Constituting an AI Commons

¶Since the industrial revolution capital has subsumed human labour and now, as it mutates into techno-capitalism, subsumes cognition; the last turn that follows is a system setting its own purposes. A clock measures how far the machine still needs us, our standing in the system as some advocate for its formal personhood. The reply is neither more regulation in the traditional sense, which moves on the electoral cycle while products ship in weeks, nor exit alone, but a governor built directly into the machine substrate and run at its pace on a recipe the cryptographic systems of the cypherpunks have demonstrated with Bitcoin: a testable rule, cheap checking, and checkers whose interests diverge and the possibility for an AI commons that hardcodes five human capacities into the internet, to constitute oneself, connect, create, transact and collectivise. Same speed, different direction.

Preface

What Is It All For?

“Whether we entrust our decisions to machines of metal, or to those machines of flesh and blood which are bureaus and vast laboratories and armies and corporations, we shall never receive the right answers to our questions unless we ask the right questions.”

— Norbert Wiener, The Human Use of Human Beings, 1950

¶A group of Silicon Valley venture capitalists and technologists, commanding the deepest pools of investment capital ever assembled, spent three election cycles translating their power and reach into a highly effective political operation; a technology super-PAC network seeded with more than a hundred million dollars to clear the road for artificial intelligence and, in the texts of its leading figures, a post-democratic order.¹ Against it a second movement has emerged, with its own lobbying groups, organised from exactly the same square mile of earth, by people who went to the same schools, worked in the same companies and raised money from the same funds, yet with a rival account of what technological progress is for and who should hold its controls. The argument over the future of humanity is being conducted inside a single zipcode. By 2026 the two had broken out into open civil war spending some eighty million dollars on congressional races with competing visions for the future of superintelligence, the data-centre rollout and the terms on which the state deals with their private AI laboratories.² Together they represent a schism in a philosophy of machines and our relationship to them that has been developing for decades and their influence on humanity extends well beyond a temporary fault line in US politics. One camp holds that the machine should be built as fast as capital allows and answer to nothing but the market; the other, that it should be paused and answer to the state. Both camps named themselves, and chose the same adjective: effective accelerationism, and the effective altruism from which the safety movement grew.1 launch of the Leading the Future super-PAC network with more than a hundred million dollars, reported by Fortune, 26 August 2025, and the Washington Post, 26 August 2025; one hundred and twenty-five million dollars raised in 2025 per its filings, reported by CNBC, 9 July 2026.2 Molly White, Tech Influence Watch, AI-industry super-PAC spending in House and Senate races, 2026 cycle (seventy-nine million dollars across sixty-eight races), as reported 5 November 2026.

¶None of this is a conspiracy, or a secret. Both sides publish manifestos, some even name enemies in formal lists, where their texts read as creeds; which is why the acceleration camp’s own popular wing calls its leadership ‘the priesthood’. Beneath the chequebooks on each side sits a philosophy held with real conviction, and given their financial weight, their reach well beyond America’s borders and their hold on the direction of technological progress, both deserve serious interrogation. The latter is in many ways the more concerning, since the remedy it reaches for is the state itself, sovereign and lab merged into a techno-totalitarian regime but the attention of the book falls mainly on the former, because it holds the throttle today, in the laboratories, the capital and the executive, and because its error is the harder to see. Its story about machine intelligence is pitched in the cadence of physics. On this account the universe is an engine for dissipating energy, and life as we know it arose because matter that organises itself burns through it more quickly, which is all the cosmos asks. Minds dissipate faster still, and capital and computation, both products of our intelligence, are when combined the latest and most efficient stage of a long and almost evolutionary process. Accelerating them is therefore not a preference but obedience to physics itself, and resistance is not merely a mistake but a kind of thermodynamic heresy. This is what has come to be called effective accelerationism, and under the memetic banner of e/acc it has gathered a following among technologists. Its believers hold it with zeal, and they include the people who decide which superhuman technologies get financed and built, as well as a Vice-President they helped put in the White House.

¶Buried inside it is some sound science, and a real insight into what capitalism and emergent AI are becoming as they fuse into a new form of techno-capitalism. Interrogated properly, however, its logic fails, and in ways that will very likely deliver the opposite of what the priesthood says it wants. The borrowed physics is not a law of nature at all but a vibe, a meme, and beneath it sits a fatalism. It is the conviction that the process is somehow divine, even inevitable, and that our part is to facilitate it or get out of the way. Two claims rest on that belief, and they fail separately. The first is that whatever wins out, in the market as in evolution, must for that reason be good. The second is that it cannot be resisted or constrained, and should not be, because progress will improve the human condition regardless.

¶The unravelling begins with a simple question that has as yet gone unanswered, and answering it is the project of this book: what is it all for? Every previous generation of engineers, economists, philosophers and constitution-writers understood this to be the first thing to ask of any form of speculative progress, not the last. Here instead it has been quietly deleted and restated as destiny, and without a stated purpose there can be no criterion against which the process could be judged to be succeeding, only a speedometer. A philosophy that erases the human question and declares itself physics is not hard-headed but naive, and in the end nihilistic, which would be a private matter if the people who hold it were not deciding the future for humanity as a whole. It is hopium in the exact sense, a drug whose active ingredient is permission to stop asking. The priesthood has its own name for it, and rather tellingly it is retardmaxxing. It has licensed a generation of founders to build without introspection, as though success had already settled what their products are for, and an entrepreneur who never asks why has left the question for the machine to answer.

¶What holds the whole edifice up is that this is inevitable. The belief is far older than the priesthood.³ Condorcet traced human improvement along a course nobody could halt, Marx gave history a motor in the means of production, and Ellul described a technique that had escaped its makers. Refuse inevitability and the structure sits considerably more awkwardly, because their engine of progress is a made thing rather than a physical law. Every layer of it, the fabs, the data centres, the power contracts, the trained models, the legal structures and the financing, was designed, sited, negotiated and signed by people, at least so far. One decision above all is shaping what it becomes. It is not the chips, the algorithms or the scale but the choice, explicit in the manifestos and implicit in the architecture, to treat the feedback that once bound the machine to human ends as friction to be removed. It is sold back to the world as a return to the natural order. Strip away the regulators and the ethicists, so the argument runs, and the market will revert to its wild and fertile primordial state. Yet there has never been a modern market without limits; property, contracts, the rule of law, the court that enforces a bargain against the stronger party and money itself, each restrain what the powerful may do. They are not the cage around the market but the lattice it grows upon. The unconstrained market is not something we once had and lost to the bureaucrats. It has only been tried once before, in miniature, in crypto’s first decade, where trading ran almost free and ended in collapse, with ordinary holders left to absorb the losses. The people proposing it now would repeat the experiment at machine speed and across the whole economy, having taken no lesson from its failure, and gambling everything on a condition already shown to fail.3 Condorcet, Sketch for a Historical Picture of the Progress of the Human Mind (1795), Tenth Epoch; Karl Marx, Preface to A Contribution to the Critique of Political Economy (1859); Jacques Ellul, The Technological Society (1954; English translation 1964).

¶Suppose the gamble is taken and the engine runs ungoverned. A governor, in the engineer’s plain sense, is the part of a machine that measures what it is doing and feeds that measure back to correct it, so that it holds the setting its builders chose. The restraints just listed work in this way, fitted to markets by societies. Without a governor an engine runs away, output feeding input with nothing to check the loop, and a runaway has to come down somewhere. E/acc says it lands in freedom. The evidence assembled in this book points the other way, to totalisation, one power taking over everything, by either of two routes. One is the state, as fear of the machine revives the oldest argument there is for Leviathan, Hobbes’s name for the sovereign to whom frightened people surrender their liberty in return for protection. The other is the market, as everything becomes convertible into a single unit of value, the cost of compute, and nothing retains the standing to refuse it. The second route is, I will argue, the more insidious, because it advertises itself in the language of liberty. Yet freedom in political theory has never meant the mere absence of limits. It is a structure negotiated so that no single hand can close around the whole, not the state’s, not the market’s, not the brilliant founder’s. Dismantling it in freedom’s name does not abolish a master. It hands control to a newer kind, one that works through the systems people now think and trade with, and the gravest danger is the hybrid in which the old physical power, with its monopoly on violence, and the new each acquire the reach of the other.

¶The clock under this argument measures dependence, how much the machine still needs us, and its fastest hand is economic. Once intelligence can be rented by the kilowatt-hour, the price of anything that can be rendered as information falls towards the cost of the energy for the computation itself, and that will be most of the modern economy. The value structures we live inside can collapse towards the cost of a model run faster than any society has ever had to re-anchor its institutions to technological change. Constraint has to be put in while the engine can still answer outward, to customers, to citizens, to law, to the grid, ultimately to us, and each year the machine needs us for less. That is the window, and it closes from the inside. What is at stake here is not simply jobs, or the automation of this or that cognitive task. It is in fact the transfer of the faculty of reasoning itself (the work of weighing, judging and deciding) from human hands into systems whose purposes are set elsewhere, held beyond our reach. Lose your job and you have lost an income, which is a loss you can still organise against. Hand over the faculty of reason and you have lost your ability to participate in society effectively.

¶Diagnosis, in any case, is cheap. The bookshelf already sags with elegant accounts of what the machine is doing to us, each ending in a closing chapter that waves towards democratic renewal without offering a mechanism that could hold a system at its own speed, and be trusted by strangers.⁴ The instruments that take e/acc apart, the political philosophy of freedom and constraint on one side and cybernetics on the other, are not only critical instruments. Followed to their conclusions they point towards something we can engineer: systems steered towards a goal people have chosen, in which engine and governor are not adversaries but one machine. The governor is not a brake fitted by the fearful. It is the difference between an engine that runs effectively and an engine that shakes itself to pieces. Purposive accelerationism is, on this argument, the form of acceleration the record favours. Whatever has lasted longest, from Watt’s mills to the constitutional republics, was built with one.4 the nearest exception is Andrew Feenberg, Questioning Technology (Routledge, 1999), with Transforming Technology (Oxford University Press, 2002) for democratic rationalisation.

¶So who then chooses the heading, and who governs the governor? The answer is a structure, not a person, a constitution rather than a wise ruler: rights that set the course, tests that check it holds, with the freedom to object, leave or start again if it drifts, and chosen ties strong enough that people stay to argue over what it is all for. They will have to hold not just people but machines as well. That does not require a theory of mind for AI, or a decision on whether a machine could ever be an agent in any formal sense and not merely the technical one. It requires only that we treat them as behaving like agents, because we need to design the systems that bind them to our society so that they stay answerable to us within it. Democracy as we know it cannot do this alone. A vote every few years never reaches the defaults written into a model, binds states while sparing the companies behind them, and gives the losing side no way out. The materials for building such a constitution already exist. Scattered across two decades of cryptographic engineering, some of it famous and some of it infamous, are components that can do a constitution’s work: ways of making scarcity, identity, property, membership and exit hold as facts of mathematics rather than grants of an authority. They were built, mostly, by the cypherpunks, the activists who from the early 1990s made the first privacy tools, and by their heirs, whose politics have failed that project in almost every way but one. It is the parts we need, not the politics. Assembled differently they are not a libertarian utopia. They are what you reach for when the ordinary structures fail, the materials from which constraint can be rebuilt at the only level the machine respects, inside its own systems, at its own speed. The people who will do that assembling I shall call the constitutionalists. Not a party or profession, but anyone, likely first among them the citizens this machine is about to displace, who takes up the writing of machine-readable rules as the alternative to being dominated. Those who would govern as monarchs already have a constitution of their own, the shareholder and the market in their most extreme form. The constitutionalist’s answer is rules written in the open, plural and revisable, rather than owned.

¶Writing a book about a thing that does not yet exist is a particular kind of act. A constitution does not describe a republic already standing. It brings one into being, the way rules make a game, and this book is written in that spirit. Nick Land, the English continental philosopher whose writing of the 1990s the priesthood treats as scripture, had with his collaborators a name for a fiction that makes itself real by being believed: hyperstition. In that sense this book is a hyperstitional act. The e/acc faithful understand this mechanism perfectly, and their texts are engineered to make their future feel already settled. The mechanism, however, is symmetrical, and it will serve a constitution as readily as a prophecy. The e/acc version contradicts itself at almost every turn. It says the future is fixed by physics and then asks for believers to bring it about, it announces a post-human future in one text and a love of human freedom in the next. Most critics read that as confusion and wonder how such an inconsistent doctrine has travelled so far, a question this book answers across its arc: that in fact the contradiction is a feature of the doctrine, not a bug.

¶A world without freedom is treated here as a default not a destiny, the outcome current dynamics are likely to deliver if no governor is built, and one we still have time to refuse. By freedom I mean something narrower than simply being free to do as you please: that no one should have the power to dominate you, whether a person, a firm or a machine, even if they never choose to use it. The guarantee is stated as a refusal, but what it protects is a set of capacities: to make, join and leave the arrangements we live under. A person’s standing to weigh, choose, contest and begin needs no defence here, and the priesthood, on its own terms of freedom, cannot deny it. I have tried to write in the temper of a constitution rather than a manifesto. A manifesto declares an end and rushes you towards it. Its mode is certainty and its relationship to you is one of recruitment. A constitution installs constraint, distributes power against itself, and keeps the end open, to be debated and revised and pursued by plural hands. It must do that through concrete rules. What follows are structures rather than outcomes, and my confidence is only in the necessity of the governor, never in a particular destination. Nobody can know with any surety which way the runaway lands, or whether building a governor will hold. Only that the attempt raises the odds of a plurality of possible futures, and of people, you, me and the generations after us, keeping the capacities that make us free to fully participate in them and to benefit in earnest from the progress they bring. Declining to try keeps the lottery open, with the ticket bought by a few in everyone’s name. For the book, only one absolute line is held: that no entity, however incorporated, should act without a natural person to answer for it, and any form of legal personhood for an artificial agent is the plainest way across it. The full case is made later in the book. Beneath that line, and more open to revision, sits a practical floor. Wherever a machine acts on the physical world, a person should stay in the loop, at a minimum as a dependency it cannot route around, because a system that still needs human hands still has to answer to them.

¶I appreciate that reading a book of some hundred and ten thousand words is a commitment, and that even listening to it runs to eight hours. But this is a complex subject, and that makes for a serious project. To the few who stay the course: I hope to meet you on the other side, to build a more purposive future together.

¶Finally, and importantly, I am not against acceleration. I am invested in it, literally and figuratively. I have spent thirteen years inside the system I am describing. I founded one of the first venture funds dedicated to many of the technologies I will critique. I helped deploy hundreds of millions into their acceleration. I have signed the cheques, sat on the cap tables, and sometimes profited from the very dynamics whose ungoverned form I am warning against. What follows is the argument of a participant who wants the engine to keep running, addressed to the only audience that can act in time: the people building it, and the bright young things about to be displaced by it — the constitutionalists I mean to arm — who between them hold the materials of its reply.

¶Same speed. Different direction.

Introduction

An Engine Without a Governor

“Men will set the goals and supply the motivations, of course, at least in the early years.”⁵5 J. C. R. Licklider, “Man-Computer Symbiosis”, IRE Transactions on Human Factors in Electronics HFE-1, March 1960, section 4.

— J. C. R. Licklider, “Man-Computer Symbiosis”, 1960

¶The scale, at least, everyone can see. The build-out now underway around artificial intelligence is, in absolute terms, the largest concentration of private capital in the history of enterprise. Trillions of dollars are moving into a single technical programme: the chips, the data centres to house them, and, increasingly the binding constraint, the electricity to run them. Firms at the frontier no longer plan their futures in units of engineers or customers. They plan in gigawatts. Power-purchase agreements have become strategic weapons, and decommissioned nuclear plants are being restarted under exclusive contract to single buyers, Three Mile Island among them, revived for Microsoft on a twenty-year deal signed in September 2024. Asked what stands between them and their ambitions, the men running these companies do not say talent, or capital, or even compute. They say energy, the raw thermodynamic input, as though the economy were being re-founded on the only currency physics accepts. Sam Altman has said that the cost of intelligence will converge over time on the cost of energy. On his account, however cheap the chips become, nobody can discount an electron. Whether that convergence is real remains an open question. But the conviction tells you what the buyers believe. Commentary on all this divides, predictably, into awe and alarm, and both camps share an assumption: that the scale is the remarkable thing. We have built at civilisational scale before — railways, electrification, the post-war highway systems, the internet itself — and the pattern of those build-outs is instructive where today’s departs from it. What is remarkable about the machine being assembled now is not what has been put into it. It is what has been left out.

¶To see what is missing, you have to know what was once there. In 1788 James Watt faced a problem that had nothing to do with making his steam engine more powerful and everything to do with making it usable. An engine’s load and its steam supply both vary. Left alone, the imbalance compounds, so that the engine either slows towards stalling or races towards destruction. Watt’s solution, adapted from a device millers already used on their grindstones, was two iron balls mounted on a spindle driven by the engine’s own output shaft. As the engine speeds up the balls whirl faster and swing outward, and their swing, through a linkage, narrows the throttle valve and admits less steam. As the engine slows the balls drop, the valve opens, and more steam flows. No operator, no schedule of permitted speeds, no inspector: a loop of iron in which a measure of the engine’s own output has been bent back to govern its input.

¶It was called the centrifugal governor, and it is in a strict sense the founding artefact of the modern world. What it added was not power but answerability. From that point a machine could be held, mechanically, to a setting its builders had chosen. Before it, a steam engine was a demonstration. After, an engine could be trusted with a purpose that remained at every moment its owner’s and never its own, which is what turned combustion into industry, because steam could now drive looms, mills and pumps unattended. In 1868 James Clerk Maxwell, taking time away from electromagnetism, published a paper titled simply On Governors, working out the mathematics of when such feedback holds steady and when it oscillates out of control. Control theory, on which every aircraft, nuclear reactor and rocket now depends, descends from that paper. From the beginning the deepest minds understood that the governor was not an accessory to the engine but the condition of the engine being effective. An engine without a governor does not run free. It runs away. Runaway is a term of art rather than a metaphor: the condition in which output feeds input with nothing to bend the loop back, so that the machine accelerates until it tears itself apart, along with the mill and anyone standing in the way. The choice was never between a governed engine and a free one. It was between a governed engine and an explosive with a delay.

¶The term governor descends from the Latin gubernator, which renders the Greek kybernetes, the helmsman at the steering oar. Who holds the ship on course through forces he does not command, the wind, the current and the vessel’s own momentum, and he could not survive by opposing them head-on. He wants the wind, and steering is what turns it into progress. When, in 1948, Norbert Wiener needed a name for his new science of purposive machines, systems that hold a course by comparing what they are doing against what they are supposed to be doing and correcting the difference, he reached past the steam engine all the way back to the steersman and called it cybernetics. Nearly three thousand years runs through helmsman, governor and cybernetics alike, and it is still at work in the industry’s own stack, where Kubernetes, the software that steers fleets of computing containers, is the Greek word verbatim. To govern is to steer. The question underneath everything that follows is who, or what, is at the oar today — and what to make of a school of thought, now commanding the largest capital formation in history, which holds that a hand on the oar is itself the problem: that the current knows where it is going, and steering is drag.

¶That school has its temples in the American technology industry, and its leadership is the priesthood met at the start of this book. Its texts are creeds, its mode is liturgy, and its anathemas are pronounced with theological confidence. Two men will stand for the movement here, as its most consequential members and its most explicit. The first is Marc Andreessen: co-author of the first mass-market web browser, co-founder of one of the largest venture firms of the software era, a16z, and author, in October 2023, of The Techno-Optimist Manifesto, the closest thing the movement has to a catechism. It names its enemies in a formal list, which is itself instructive: sustainability, tech ethics, risk management, the precautionary principle. Every entry is an institutional form, imperfect perhaps, by which modern societies bend a technology’s output back towards its input. And it rests its optimism on energy, which it calls life itself, and names among its patron saints the founders of e/acc, whose borrowed thermodynamics has the universe conveniently wanting what they want. In genre it is a hymn rather than a brief, and hymns are not meant to be examined. They are meant to be sung.

¶The second is Peter Thiel, who built PayPal, took from his Stanford teacher René Girard a theology of desire and sacrifice he has never stopped applying, wrote Facebook’s first outside cheque, took Palantir into the security state, and supplies the movement’s temperament and its candour. In 2009, in an essay for the libertarian journal Cato Unbound, he wrote the sentence to which everything that follows is an extended response: “I no longer believe that freedom and democracy are compatible.”⁶ The extension of the franchise had, he argued, made democratic majorities hostile to the libertarian project, so freedom would have to be pursued where votes cannot reach — in technologies that “create a new space for freedom” beyond the reach of democratic politics.⁷ He was describing his own loss of faith in elections as a route to liberty, not proposing to disenfranchise anyone, and when the essay drew fire he added that no class of people should lose the ballot, only that he held little hope it would make things better. Thiel matters more to this argument than a thousand vaguer enthusiasts because he has written his premises down. Where others gesture, he states: when freedom and collective self-rule come apart, it is democracy that should give way. That is no retreat from politics but the choice of a different kind, in which whoever builds the new spaces writes their rules. The proposition has serious money behind it, and everything turns on the work the word freedom is performing in his creed. The priesthood is formidable. Its argument is that technology has been the great engine of human flourishing and that halting it would cost more than it saves. By the measures that decide what gets built, capital deployed, political money spent and the absence of any rival in speed, it is winning. Answering it requires knowing where its ideas come from, which is somewhere older and stranger than a venture firm’s marketing department.6 Peter Thiel, “The Education of a Libertarian”, Cato Unbound, 13 April 2009.7 Thiel, above.

¶Accelerationism’s roots run back to French philosophy of the 1970s, and behind that to the Italian Futurists, who in 1909 declared a roaring motorcar more beautiful than any statue and went on to write the founding manifesto of fascism. But it took its modern form not in California but in Warwick, England, in the 1990s, inside a renegade academic group called the Cybernetic Culture Research Unit. Founded in 1995 by the cybernetic feminist Sadie Plant, it was inherited and then radicalised by Nick Land, whose feverish theory-fictions broke with the academy and seeded accelerationism as we now know it. Land’s thought is radical in the strict sense, and no summary survives contact with his prose, which is fast, polemical, funny and frequently brilliant. At its core is this: capitalism, fused with computation, is not a human institution at all. Importantly, by capitalism he means the profit-motivated system of private ownership, not the mere existence of markets. For him it is an inhuman process, self-exciting, self-expanding and indifferent, which uses human beings the way a wasp uses a caterpillar, as a living host for its young. Humanity is not the protagonist of history but its larval stage, and the process now accelerating through our economies is the thing we are in service to. Where a humanist recoils, Land’s writing exhibits something closer to worship.

¶This is the Valley’s intellectual inheritance: not Adam Smith, not even Ayn Rand, but a 1990s philosophy of capital as an alien intelligence bootstrapping itself. Land put it at its most extreme when he called capitalism a kind of artificial intelligence, “an invasion from the future by an artificial intelligent space that must assemble itself entirely from its enemy’s resources”.⁸ On this reading the machine intelligence now emerging is no novelty at all but capital becoming literal, finding in silicon the substrate it had until now borrowed from us. Land’s claim, however, mistakes a name for an agent. Capitalism is a process built from arrangements — property, contract, price, credit — much as the weather is wind and rain, and there is nobody inside it to want anything. The horror is real but mislocated. It needs no alien mind using us, only a process automated from end to end, each result feeding the next, that runs with no will at all.8 Nick Land, “Machinic Desire”, Textual Practice 7, 1993.

¶It was from here the inheritance forked. On the left, Alex Williams and Nick Srnicek argued in their 2013 #Accelerate manifesto, the founding text of what is now called left accelerationism or l/acc, that the machine should be accelerated and captured, that the technologies capital builds could be repurposed by political mastery for human emancipation. This version kept the question what is it for? at the centre. The Valley’s variant, effective accelerationism, whose creed was authored principally by the physicist-founder Guillaume Verdon, performed the opposite surgery on each parent. From Land it kept the sovereign process — capital as the thing that answers to nothing while everything answers to it — and discarded the dark honesty about where that process leaves us. From l/acc it kept the speed and discarded the emancipatory politics. Its authors wanted automation to win back people’s time, and in their 2015 book Inventing the Future they asked for a shorter working week and a universal basic income, demands addressed above all to the state. Its popular cousin, Aaron Bastani’s Fully Automated Luxury Communism (2019), went further and recast the communist ideal for an age of machine abundance. What remains, once you have removed both the destination and the question, is a throttle. E/acc is the doctrine of throttle: pure acceleration, cosmically licensed, with the steering reclassified as drag.

¶The family has produced one serious internal dissent. In late 2023 Vitalik Buterin published the case for d/acc, its d standing at once for defensive, decentralised, democratic and differential acceleration.⁹ Buterin is the nearest thing crypto has to a resident philosopher, ironically a Thiel Fellow who left university at 20 on a hundred-thousand-dollar grant to build Ethereum, and one of the few figures the priesthood cannot dismiss as a Luddite. Accelerate, he argued, but accelerate selectively, favouring the technologies that defend and decentralise over those that attack and concentrate. This was an advance, the first accelerationism native to the technology industry to admit that direction is not one variable among others but a presupposition. And it concedes the decisive premise: technology answers to values, and the only question left is whose. But selection alone is not steering. Favouring defensive technologies expresses a preference. It does not supply the reference value, the chosen heading against which a system’s behaviour is measured and corrected. D/acc fits the brakes and the airbags and asks that the passengers be heard, but the vehicle still has no steersman. The advance is real, and so is the vacancy at the oar.9 Vitalik Buterin, “My Techno-Optimism”, vitalik.eth.limo, 27 November 2023.

¶The territory of all this is usually mapped on a single axis, faster or slower, which is just the picture the priesthood needs, because on one axis every governor looks like a brake. Drawing the real map, as the Three-Axis Map figure does, requires three: speed, running from deceleration to acceleration; the governor, from absent to present; and the shape of power, from a single to plural hands. The decelerationists — degrowth, the precautionary establishment — hold the slow-and-governed region, and their position concedes the decisive ground. Moratoria have worked before, on human cloning and on germline editing, but never against a technology with a trillion-dollar industry lobbying over competing visions and a global arms race behind it with no definition a legislator could write down. The engine will not be stopped by people who do not hold the throttle, and the attempt guarantees that whoever wins on speed faces no rival there. In November 2026 the slow corner found its electorate, when voters in ten American states claimed the right to refuse a data centre.¹⁰ The priesthood holds the fast-and-ungoverned corner with overwhelming force, and drifts along the third axis as it does so, because unchecked concentrations of power do not stay plural. At the time of writing, in late 2026, China occupies fast-and-governed with the power axis closed: an acceleration with a steersman, a party-state whose helm answers inward, to its own hierarchy and to nobody else. On two of three axes China and I agree, and we part on the third, because a governor answering to a single hand is itself a totalisation, and the oldest. Among the democracies, the European Union comes nearest to governed and plural, and pays for it in speed; Britain for now runs fast and lightly governed, and the United States fast at the frontier and contested at the grid. The left accelerationists stand nearest the corner this book argues for, fast and steered, but they fix the ends in advance and lean on the state to keep them.10 Ballotpedia, “Voters in ten states will vote on data center-related ballot measures November 3”, 3 November 2026; results as compiled by Ballotpedia and CNBC, 4 November 2026.

Figure · The three-axis map

Speed is the axis everyone argues about. The other two axes decide.

Drag to turn the map. Each position is a point in three dimensions: how fast it would go, whether a governor holds it to a reference value, and whether power is a monopoly or distributed. Flatten it to any two axes, or pick a position to see where it sits and why.

Drag to turn · arrow keys work too

Camps

Jurisdictions

Where it sits

Governed means answerable to someone other than the party restrained. Each point is where a camp or power stands at the time of writing, in late 2026, or in principle, not where it is heading. Placements are judgements, not measurements; the Great Game tracker follows named people, firms and states over time.

¶The map filled in during one week of September 2026. OpenAI called for mandatory federal safeguards. Anthropic’s chief executive, Dario Amodei, published a plan to pace the frontier and asked Washington for an antitrust waiver so that the foremost companies of the democracies could agree limits on the rate of their own progress, and Sam Altman endorsed him within hours. David Sacks, until March of that year the White House’s adviser on artificial intelligence, told them to go ahead, since the two firms already held “a duopoly on frontier intelligence”, and to stop pretending they needed a waiver or an approval process that superseded product liability.¹¹ A moratorium had its advocates both in Congress and, days earlier, in a letter from more than seventy UK MPs and peers asking for a ban on superintelligence. The leading labs had moved to fast and held in two hands, governed only by a throttle they would set themselves, with the state as guarantor. Their critics held fast and plural, with liability and the customer as the only governor they would admit. The movement’s popular wing, e/acc, went further, meeting the labs’ pacing plan with a call for decentralised AI that anyone could run without permission, an open stack of models, tools and local inference. That hints at a direction, and it is an improvement I endorse, but on its own it is a topology. It says where the weights sit, not what they answer to, and a model on your own machine can carry an objective you cannot inspect as easily as one on a frontier lab’s server. By the end of the month six of the industry’s leaders had signed a voluntary accord at the White House, binding on no one, and just six weeks later voters backed a hostile legislature against a president with a veto, leaving a bill that would stop the work on pain of prison as the only federal alternative. Each offer puts a single party in charge of the speed.11 David Sacks, post on X, 13 September 2026 (x.com/DavidSacks/status/2098973625252708460).

¶The corner that stands empty, and the subject of this book, is fast, governed and plural: full speed, with a measure of the output bent back by design, at machine speed, to hold the machine to humanly chosen and humanly revisable ends. This is the position the book’s title names, purposive accelerationism, or p/acc in the family’s own naming convention. Purposive, because what distinguishes it from e/acc is not less throttle but the presence of a destination. Accelerationism, because unlike the decelerationists it wants the engine run at full speed and the technology built. That corner has never been empty of people. Engineers, cryptographers, constitutional cypherpunks and Buterin himself have all stood in it. However, it has been empty of architecture. Its occupants have had instincts, tools and ways out, the means to leave a system rather than change it. What they have lacked is a constitution, meaning a worked-out account of what the governor is, what reference value it holds, what materials it is built from, and why it binds. Supplying that is the work of this book. Its answer is not a charter asking governments or firms for anything. It is a set of rights built into the internet itself and, importantly, into the emerging AI stack, a cognitive commons no single owner can close, with ways for those living under it to steer and tests to measure its degree. They are gathered on a page in the appendix: the reference value, the criteria any governor must meet, the rights, the tests that check them, the provision and the rulebook, and the one red line never to be crossed, for any reader who wants to keep score as the argument builds. The engineering, which primitives in which arrangements at which parameters, belongs to founders and the market, so that nothing argued here dates with any particular configuration. That matters more than it sounds, because the materials themselves carry a theoretical expiry date. The public-key cryptography beneath every key and ledger we have pointed to in these pages is scheduled for retirement, deprecated by the body that standardises it from 2030 and disallowed from 2035, because the expectation is a large enough quantum computer breaks it. Its successors however are not waiting on that machine; they are ordinary mathematics, chosen in an open competition that ran for eight years, and already published. So the architecture proposed here has to do one thing the cypherpunks never had to: carry a constitution across the replacement of its own electronic foundations, with no party in charge, on a deadline set by physics. That is not an objection to the project but its first test, and a protocol nobody owns is the only kind whose passage everyone can watch.

¶My own position needs stating before going further. Alongside a decade of investing in crypto, I was an early advocate of something called platform cooperativism, a movement that set out to give the people who create a platform’s value a share in owning and governing it. The claim made here for crypto is accordingly a narrow one. Co-operatives, mutuals, unions and commons have held shared goods by construction for centuries and still do. What crypto adds is the ability to hold such things across borders, at the speed machines now work, without anyone’s permission, which none of the older forms can manage.

¶Two sets of instruments are needed for our work here, one technical and one philosophical. The technical set comes from cybernetics: Wiener’s account of purposive systems, widened by Gregory Bateson into an account of minds that extend beyond any one skull, and hardened by Stafford Beer into an architecture for institutions that can hold a course. That tradition named the governor half a century before the priesthood deleted it. The philosophical set reaches further back still. Isaiah Berlin and Amartya Sen, writing when freedom was at its most contested in modern history, supply the anatomy of what it consists of, and catch the priesthood invoking one kind of liberty while its prescriptions serve only the other. Hume set the bar the priesthood must clear, by showing that no quantity of is yields an ought, and Moore devised the test that no natural definition of the good survives. Both blades fall across e/acc’s thermodynamic catechism. And finally Arendt supplies what the reference value must in the end protect, which is plurality. The witness I most look forward to calling is Friedrich Hayek, the priesthood’s own house god. He understood the price system as a feedback device, a governor for allocation though never for direction, and his own writing is a sustained argument that it works only inside constraint. He will testify, against his admirers, that the wall they are demolishing is load-bearing.

¶My method is to take the strongest version of the case against constraint, stronger than its proponents might usually state it, grant everything in it that is true, and then show that the single move that matters, the deletion of the governor, does not follow from any of it. A defeat on home ground, by the movement’s own first principles, shows that an acceleration which keeps its governor is not a timid compromise but the only version of the project that is coherent, and truer to what the priesthood says it wants. Nobody has ever been argued out of a religion by being told it was stupid. They have occasionally been shown, from inside, that their god requires something other than what the priests demand.

¶The reason to do this now, rather than watch another decade of evidence accumulate, is a fact about infrastructure that every engineer knows and every politician learns too late: constraints are cheap to design in and ruinous to retrofit. The internet’s original sins — no native identity, no native money, no native provenance — were each a decade of papers away from being design choices. They are now structural absences that entire surveillance and platform economies have calcified around, and some may never be corrected. The machine being assembled today at gigawatt scale is still in development. Its protocols are still being written, its settlement layers are still contested, its governance, where there is any, is still molten. For a short time, what it optimises for is a design question rather than an archaeological one. The governor can be specified now, by people who understand the substrate, or excavated later from the rubble. So the case must be heard before it is answered. The strongest argument against everything I propose — the case for removing the governor outright, made in earnest by the most capitalised movement on earth — deserves its day at full strength. It begins, as the best arguments do, where everyone notionally stands: with the cause of freedom.

Subsumption

¶A governor can be put back only if the case for going without one has first been heard at full strength, so this section begins there. It grants the argument against constraint everything that is true in it, and then asks what that argument leaves unsaid: who sets the ends towards which an ungoverned machine runs, and what becomes of the people whose work, craft and judgement it is built to absorb. From there it follows the engine to the places it is most likely to come down, and reads the clock that measures how long a remedy stays affordable. The verdict does not rest on any guess about what the machine intends. It rests on sums anyone can check, and on a choice that remains open only for as long as someone still has the means to make it.

Chapter 1

Against Constraint

“the end of law is not to abolish or restrain, but to preserve and enlarge freedom: for in all the states of created beings capable of laws, where there is no law, there is no freedom.”¹²12 John Locke, Second Treatise of Government (1690), chapter VI, section 57.

— John Locke, Second Treatise of Government, 1690

¶In 1820, roughly three of every four human beings alive were poor in an absolute sense: short of food in a bad season, dead before sixty, and likely to bury nearly half their children before a fifth birthday. The line on the chart that tracks this condition runs flat across ten thousand years of agriculture, empire, religion and philosophy, through Athens and the Renaissance and every moral and cultural advance our species made. Then, somewhere around the time Watt’s engines began driving English mills, it bends upwards into the shape economists call a hockey stick, so steeply that everything before it reads as a footnote to the last two centuries. Today about one person in ten lives in the old condition, on a planet carrying eight times as many of us. Life expectancy has doubled. Literacy, once the property of clerks and clergy, approaches the human default. Nothing else in the historical record resembles this.

¶In the priesthood’s telling, the engine for this was the pairing of machines with markets: technological capability compounding under the economic system we have come to call capitalism. This, in reality, is only half the picture. The deeper spring was the one Francis Bacon named 170 years before Watt, an open market not of goods but of ideas, in which observation, experiment and public reproduction displaced inherited authority. Germ theory, sanitation and vaccination then translated that market’s findings into the survival curves the chart records. Capitalism was a powerful transmission and the scientific method the engine, though historians still argue the priority of causes. What compounded was the pairing: machinery yoked to a method for finding out what is true. But the priesthood’s version, with capitalism as its sole or primary operative, is the one with the money behind it.

¶If that pairing alone produces the escape from poverty, then whatever gets in its way has a price, paid in human lives. A drug that spends three extra years in licensing kills invisibly. The power plant not built holds a village in the smoke of its dung fires. Golden Rice, the crop variety strangled when a Philippine appeals court revoked its permit in April 2024, leaves real children blind. Here, the cost of delay is harder to see than the risks it was meant to avoid, because a statistical morgue has no photographs. Marc Andreessen’s Techno-Optimist Manifesto repeats the mantra without flinching, proclaiming that any deceleration of AI will cost lives and that deaths “preventable by the AI that was prevented from existing” are “a form of murder”.¹³ The conclusion follows: accelerate at any cost, and treat every hand that reaches for the brake as an enemy not merely of progress but of humanity. Its mirror image follows just as easily, that whoever delivers the curve is humanity’s friend, owed thanks rather than questions. The Manifesto goes on to set that conclusion to choir music. Growth is the path out of every problem, technology is its engine, markets are its accelerant, and against them stands a roll of named enemies. Beneath the optimism sits the older and stranger cosmology of effective accelerationism, and its argument is simple. The universe favours whatever burns through energy fastest, and life, intelligence, technology and capitalism are the most efficient such structures yet to emerge. Speeding them up is therefore obedience to the deepest law there is, and resistance is not error but heresy, an eddy setting itself briefly against the current of a roaring river. Guillaume Verdon, a physicist and one of the movement’s founders, calls this the thermodynamic god.13 Marc Andreessen, “The Techno-Optimist Manifesto”, Andreessen Horowitz, 16 October 2023.

¶Strip the liturgy away and a genuine argument stands underneath. Freedom is good, a premise nobody easily rejects and the ground on which Locke, Mill, the abolitionists and the priesthood all stand together. Whether it outranks health, security or a fair start is disputed, and nothing in my own case depends on it: I need only that a society which loses the capacity to choose cannot then weigh what else it wants. Progress expresses freedom, because material capability is liberty in practice, and the serf freed only in law remains bound by the field. Capitalism is the precondition of progress, since of all the systems ever tried only one has compounded capability for two centuries, and it does so by harnessing dispersed knowledge and self-interest to the work of improvement. From this it seems to follow that every obstacle to capitalism’s operation is an obstacle to freedom itself, to be resisted and ideally removed: the regulator, the ethicist, the journalist, the committee, the inquiry, every institution whose function is to interpose a question between capability and deployment. And the stakes have increased, according to Nick Land, whom we met earlier, because capitalism and artificial intelligence are now fusing into a single machine, a techno-capitalism that manufactures intelligence itself, so that every year of friction is a year of the hockey stick foregone.

¶It is a powerful and seductive argument, and it is also a case for removing the governor, which works by feedback: it compares what a system is doing against what it is meant to do, and corrects the difference. Watt’s flyballs compared actual speed against a set speed. A thermostat holds heat to a set temperature. A drug regulator measures the medicine against the standard safe and effective for the patient, and a constitution tests the law against the standard compatible with the citizen’s rights. Every governor, mechanical or institutional, presupposes a reference value: a setting, chosen from outside the mechanism, that says what the system is for. Two kinds of rule are at work here. Some rules restrict something that would go on without them, the way a speed limit restricts driving. Others make the activity itself, the way the rules of football make the game and the rules of property make a market. Rules of the second kind are why there is anything to regulate at all.

¶The argument goes further than loosening the setting and declares any reference value for techno-capitalism illegitimate, because the question — supposed to do what? — is exactly the one it wants removed, on the fear that any answer whatsoever empowers someone to slow the engine. Each enemy on the Manifesto’s list, from sustainability to the precautionary principle, is a way of deciding who is allowed to ask what the engine is for. Many of them are imperfect institutions and some are poor ones. The precautionary principle earns the criticism it gets: in its strong formulations it demands proof of safety before action, which is a standard no technology in history has met, and it has been used to block vaccines, nuclear power and modified crops at measurable cost in lives. The Manifesto moves from the fair criticism that this instrument might be badly designed to the claim that the function it serves is illegitimate, and those are different claims. A bad thermostat is an argument for a better one, never against the idea of a set temperature. So the priesthood identifies the obstructions, and its argument says freedom requires ripping them, all of them, away. The case rests on four words — freedom, progress, capitalism and all — and not one of them survives inspection.

¶In 1958 Isaiah Berlin, who had watched both of the twentieth century’s totalitarianisms from close quarters and made liberty his life’s subject of study, pointed out that the word freedom names two different things. Negative liberty is freedom from, the absence of interference, the space in which nobody may coerce me. Positive liberty is freedom to, the presence of capacity, the ability to actually author a life. Debate usually treats the two as interchangeable, which is why the substitution often goes unnoticed, yet a starving man on an empty plain has all the negative liberty in the world. Amartya Sen turned the second idea of freedom into the measure now standard in development economics: a person’s freedom consists in her capabilities, the lives she is actually able to lead, and a society’s in the breadth of its people’s capability sets, the full range of options open to each of them. Read either way, a society that narrows what its people can do has made them less free. Berlin himself trusted the negative kind more, having watched positive liberty, administered by states, become tyranny’s alibi. The move he feared had been supplied by Rousseau. If there is a true self that wills the good and a lower self that mistakes it, then whoever knows what the true self wills may compel the lower self towards it and call the result liberation. Anyone refusing, Rousseau wrote, “will be forced to be free”: six words that have stood ever since for the totalitarian temptation, domination rebranded as emancipation.¹⁴ A governor built on that claim would be what he called the general will, running in a server rack. This warning applies to governors as such, and whatever I defend in this book has to satisfy it.14 Jean-Jacques Rousseau, The Social Contract (1762), Book I, chapter 7, in G. D. H. Cole’s translation.

¶Hold the hockey stick we looked at earlier against the distinction and it splits. The escape from poverty is a story about positive liberty. The child who survives, the woman who reads, the family whose horizon extends past the next harvest: when the argument invokes the moral force of two centuries of flourishing, capability is what it points at. But its prescription serves only the negative kind. Strike out the regulator, the ethicist and the planning inquiry, and each deletion widens the space of non-interference directly, and often expands capability too. The trouble is the word often. Sometimes the two freedoms move together, as with the unshackled drug that saves ten thousand. Sometimes they come apart, as with thalidomide, which one unhurried regulator, the FDA reviewer Frances Oldham Kelsey, kept out of American pharmacies. The doctrine has no way of telling the cases apart, because sorting them requires the question it has ruled illegitimate: whose capabilities does this particular removal enlarge? That question is a reference value, the setting a governor holds to and can never supply for itself. A person installs it, and the mechanism only compares and corrects against it. So the Manifesto rests its moral case on capability while refusing, on principle, ever to be measured against it.

¶The second word is progress, and here the equivocation is built into the grammar. Progress means movement towards something. The word carries no meaning without a declared destination, a purpose, supplied from outside the moving thing. Take the destination away and what remains is simply change: more, without necessarily better. The catechism uses the word as though more technology, faster, were a destination rather than a rate. A tumour does more, faster and with mounting efficiency. So do a wildfire and a rumour. None of them is going anywhere a human being has reason to cheer, because the standard of advance needs to be set by us, from outside the process, or it is not set at all. Movement without a destination is not an advance but escape.

¶There is a second slippage inside the same word, somewhere between liberalism and progressivism. Liberalism is a claim about liberty, that each person’s space to author a life should be maximised, and it carries no promise that tomorrow improves on today, which is why a liberal can be temperamentally conservative: the farmer who wants nobody telling him what to grow. Progressivism is the separate claim that the future will be better than the past, that the arrow of time is also an arrow of value. The priesthood borrows the authority of the first while asserting the second. The tradition I draw on is liberal, and the progress I refuse is progressivist, because a claim about liberty can be tested against a life, and a claim that the future somehow must be better can only be taken on faith.

¶The argument might still stand if the third word, capitalism, held it up. If capitalism were good in itself then more capitalism would be progress by definition, and the destination would be supplied. Here sits the subtlest slip of the four: an instrument quietly promoted into an end. The honest case for capitalism, the one Adam Smith made and the one the hockey stick supports, is instrumental. Markets are good because they deliver flourishing. Dispersed knowledge beats central plans, competition disciplines waste, the proverbial baker’s self-interest feeds the town. Every clause measures the instrument against a standard outside it, and stands ready to find it wanting case by case. That readiness is what makes it an argument rather than a creed. Somewhere in the priesthood’s catechism the relation flips. The standard drops out while the instrument remains, claims begin to be tested against the market’s own operation, and it serves the market is taken to settle a question rather than open one. Put to the doctrine a constraint its own audience would concede improved human lives at some cost to capital: the child-labour law, the eight-hour day, breathable city air. An honest instrumentalist takes such things case by case. The catechism cannot, because a doctrine whose enemies are whole categories has ruled out sorting in advance. An instrument that may never be checked against human flourishing no longer protects it. It serves the engine instead, and blindly.

¶Grant all three premises and the conclusion still does not follow, because it smuggles in a fourth word: all. From the claim that some limits obstruct, the priesthood concludes that they all must go, yet nothing in them licenses the leap. A cylinder wall is a constraint on an explosion, the most literal possible interference with the fire’s freedom to expand in every direction. It is also the only reason the blast drives a piston rather than scattering the machine across the shop floor. The wall is why there is power rather than debris. That is to say, the whole craft of engineering, like that of constitution-writing, lies in telling load-bearing walls from genuine obstructions, and that is exactly the distinction the word all erases. Knock the walls out in the name of freeing the fire and you do not get a faster engine. You get a loud bang.

¶A careful accelerationist could concede all four points and rebuild the case on capability: name a destination, treat markets as a means rather than an end, and keep the constraints that make growth possible. He would arrive somewhere close to the position I defend. What blocks that repair is the cosmology, an argument for acceleration drawn from a supposed law of nature, and its faults run deeper. The first is that the destination is still missing, and this version merely hides the absence better. Physics supplies the direction, so the claim goes: the universe runs down towards disorder, which science calls entropy, structures that speed that descent are favoured, intelligence being one such structure, and accelerating it therefore carries the endorsement of thermodynamics itself. The physics invoked is real. Ilya Prigogine won the 1977 Nobel Prize for explaining what he called dissipative structures — a whirlpool, a flame, a living cell — which stay far from equilibrium, the dead calm in which nothing moves, by burning through whatever flows into them. Jeremy England took the work further, showing that clumps of atoms bathed in an energy flow tend over time to rearrange themselves so as to absorb and spend it more effectively, which is adaptation without genetics. And Karl Friston’s free-energy principle points the other way: it describes living systems persisting by minimising surprise, which in practice means staying within the narrow range of conditions they can live in, which makes it a rule of conservation rather than of burn. Even the science nearest the priesthood’s own casts intelligence as something that survives by governing itself. The mistake arrives in the sentence that slides in after the physics: and therefore we ought to accelerate. An ought ranks states of the world, and no description of how things are, or have been, or tend to be, contains a ranking. The world as it stands is not always as it should be, which is the function of “should”. Thermodynamics is a forecast rather than a commandment. It tells you what matter will do under an energy flow and is silent on whether to cheer.

¶If the telos, the end a process aims at, is entropy, then the heat death is the destination, and we win when everything including us has been smoothed into the same lukewarm haze at the end of time. Make the goal complexity that feeds on energy flows rather than exhausting them, and the problem only moves one level up. Nature’s portfolio of dissipative structures includes the hurricane, the wildfire and the plague, each a magnificent exploiter of gradients, the differences in heat or pressure that make things move, and physics does not rank them by preference. The ranking — intelligence over weather, civilisation over fire — is made by us, and the universe is then dressed up to look as though it made the choice first. The second fault is entangled with the first, which found the destination missing, for here one is present, disguised as a discovery. The direction was chosen, largely by men in boardrooms within the same few square miles of California, holding portfolios much like mine, and the physics was draped over the choice afterwards. The rhetoric of the movement is one long imperative — build, grow, multiply, ascend — issued by a doctrine whose official metaphysics contains no imperatives at all. The human question has not been deleted, only renamed physics.

¶The grounds for refuting all this were settled long before anyone built a data centre. Hume set down the first in the Treatise of Human Nature, in a passage that has lost none of its edge. In every system of morality he had encountered, he observed, the authors proceed in the ordinary way of reasoning, in is and is not, and then, imperceptibly, the propositions begin arriving joined by an ought. The e/acc catechism performs that sleight of hand at a cosmological scale: a hundred true sentences about entropy, and then, somehow, imperceptibly so build. Moore supplied the second in 1903. He named the naturalistic fallacy, the attempt to define goodness as some natural property — pleasure, evolutionary fitness or, for the priesthood, dissipative efficiency — and devised a test for it. Where a definition holds, the question dies. Ask: this system dissipates energy at a magnificent rate, but is it good? The question stays alive, open and worth arguing about, which means dissipation and goodness are not the same property and the cosmology has been trading on the confusion.

¶The third fault is the appeal to nature, and it falls down on its own evidence. Nature endorses without favourites: the immune system and the plague, the harvest and the locust, photosynthesis and the heat death. A philosophy that reads moral instruction off the universe’s tendencies must take the whole syllabus, cold iron and event horizons included. The priesthood reads selectively — life, intelligence, growth — and the selection principle is its own: human, normative and undeclared. Darwin’s natural selection has no goal, being a blind record of what replicates, with neither direction nor design. The priesthood reads it as though the world had an inbuilt end, as the acorn has the oak, so that whatever the market favours becomes what is meant to prevail. Darwin emptied the process of purpose, and the priesthood put one that served it right back in.

¶None of this machinery is new. Its direct ancestor is Herbert Spencer, the Victorian railway engineer turned philosopher, who defined better as more evolved and was one of Moore’s own exhibits. He had the politics before he had the science: the laissez-faire ethics of Social Statics appeared in 1851, eight years ahead of On the Origin of Species. When Darwin arrived, Spencer coined “survival of the fittest”, draped the new biology over the old programme, and read the result back as nature’s instruction, with the poor as the unfit in mid-elimination and charity as interference with the species’ refinement. The American business class took it as gospel, and Andrew Carnegie, the Scottish-born steel magnate who became one of the richest men of his age, wrote of the light that came with Spencer. It found its congregation among the era’s winners for the same reason e/acc now fills the Valley’s pews. It told them their success was not merely permitted but cosmically ordained: the dead were physics and the profits were destiny. Then and now the structure is identical: a real science, a bridge based on fallacy, and a payload holding that the strong owe nothing and the brake is sacrilege. Effective accelerationism is to thermodynamics what Social Darwinism was to evolution, and the chronology repeats too: the politics came first and the physics was retrofitted. Darwin’s science survived the collapse of the ethics that hitched a ride on it. The thermodynamics will survive e/acc the same way.

¶There remains the empirical wing of the case, the free-market reply that unconstrained markets simply work and two centuries of the record prove it. Inside it sits the priesthood’s most reasonable argument: the market is itself a governor, in this view the ultimate one, so nothing else needs building. This is half right, and the decisive witness on both counts is the priesthood’s own house god. Hayek is regularly cited wherever a regulator is being shown the door, and the use made of him often inverts what he wrote. His case for the market was epistemic, an argument about what can be known rather than about what is right. Knowledge of circumstances, of time and place, is dispersed across millions of minds and cannot be assembled by any central authority. The price system is the only known instrument that can aggregate it, a discovery procedure for facts no committee could collect. Prices report scarcity, profits reward the answering of need, losses punish waste, and millions of plans are corrected daily against a signal none of them controls. That is a governor in the fullest sense: distributed, fast and running at the speed of the thing it governs.

¶What Hayek discovered, though, is narrower and stronger than is often made of it. His insight concerns dispersed knowledge and its coordination, and the price system is one mechanism for that work, tuned to knowledge of a particular kind: tacit, local, held in circumstances its holder could never fully articulate. The scientific method coordinates dispersed knowledge of another kind, made of articulated claims tested by reproduction and criticism and rewarded in reputation rather than ownership. Peer review, replication and public criticism are a governor too, holding a body of claims against a reference value, what survives testing, that no individual scientist may set. The genus is the process of discovery, prices and peer review are two of its species, and capitalism owns only one of them.

¶On the narrow claim the priesthood and I agree. A market operating within its constitutional casing is among the most successful governors ever built, something I have spent a career inside. Hayek also wrote two conditions into each step of the argument, and both cut against the people who misquote him. The first is that the discovery procedure runs only inside a framework of rules — property, contract, fraud and its punishment, money worth holding — and no market produced those rules itself. They cannot be bought into existence, because buying presupposes them. Every functioning market in history inherited its framework from outside itself, from law, custom and constitution, and people, acting together, chose it. Hayek spent his last decades writing about almost nothing else. The man invoked against constitutionalism wrote three volumes of constitutional design, and ended by proposing to take money out of the hands of governments because they could not be trusted to hold to a rule, even one of their own making. He defended freedom under rules. His lifelong target was arbitrary power, discretion unbound by law, whoever held it. He also held that dispersed private ownership was itself a check, the material condition that stops any single authority commanding everything at once, an argument that bears as much on a billionaire founder holding the whole as on a ministry, and one his admirers quote less often.

¶The second condition returns us to the reference value. The price system governs allocation and never direction. It answers one question — how should resources flow, given what people with money want today? — and nothing else. Whatever carries no price casts no vote: the river poisoned downstream, the unborn generation, the catastrophic tail that arrives too late to bargain. A governor blind to everything unpriced is not neutral among values; it holds the priced ones, installed by default, which is how the Thames ran as an open sewer until the summer of 1858, until the smell reached the House of Commons and Parliament paid for drains within weeks. So the market is the ultimate governor of one question and no governor at all of the other. The priesthood’s programme works by collapsing the two, deleting every chosen reference value, keeping the emergent one in its place and calling the result freedom. That is what being steered by unchosen dynamics means in practice.

¶If progress requires a destination, and nature supplies none, and the market steers allocation but not direction, then the reference value must be chosen — deliberately and explicitly, in the open, by people answerable for it — and built back into the machine as a governor. Refusing to choose is not neutrality. A system of this power will be steered by something. Decline to install a reference value and the ungoverned dynamics supply one: capital growing for its own sake, the engagement metric, the pull of whatever is cheapest to optimise, or a government capturing the technology by declaring it a threat too grave for ordinary rules. That setting is chosen by no one, answerable to no one, and wears the refusal as its legitimacy. A system is purposive when it is steered by a reference value and kept correctable by humans, meaning that a setting exists which says what the system is for, and the people it serves can inspect and revise that setting while the machine runs. Watt’s governor is purposive in the small: the set speed is chosen, visible and adjustable by the millwright. Purpose and control are different properties, and a system is purposive only with both. A machine drifting aimlessly may still be steerable, while one holding a goal no human hand can revise is not.

¶A system is human-aligned only when its reference value preserves the conditions for plural human authorship. Any thicker definition, alignment with welfare or with virtue, collapses into the question whose values? This account rests on two features of the human condition that the political philosopher Hannah Arendt identified. Plurality: the world is inhabited by many distinct people, irreducibly different in perspective, none entitled to stand for all. And natality: each new arrival is a capacity to begin something unforeseen, so the future stays structurally open. A human-aligned system protects the standing of every person to hold a conception of the good, act on it, and start something new. These are the preconditions of valuing itself, and they name no winner among particular goods. Plurality is still a choice, defended as the one commitment no valuer can coherently refuse, since arguing against the preconditions of valuing is itself an exercise of them. Alignment in this sense leaves open where the engine goes and requires only that the destination remain choosable, by many hands.

¶Agency is the good, not the bare absence of interference but the capability of persons, plural and distinct, to author their own lives. Non-domination, as later chapters will explain, is its condition, because agency survives only where no power, public or private, holds another’s possibilities at its discretion. Progress is real and measurable, though only against a chosen and revisable reference value, and the choosing is a politics no one can delegate to physics. The market is a magnificent instrument, a governor in its own right, which serves agency exactly as long as its constitutional casing holds. Therefore yes, I believe we should accelerate, but we should build the governor into the machinery of the acceleration itself, at the speed of the thing it governs. This position must withstand three replies. The first is about legitimacy: who chooses the reference value, by what right, through what procedure, and against what capture? The second is about time: whether there is still any, and whether constraint can be built into systems hardening by the quarter, faster than constitutions have ever been written. The third is whether a governor, in practice, is anything more than a brake by another name.

¶The third can be answered at once, here, because as we have already shown the charge that a governor is only a brake fails on technical grounds. Watt’s governor never slowed the steam engine. It allowed it to drive delicate machinery at loads and revolutions that would have wrecked an ungoverned one, so the device made speed usable rather than holding it back. Where failure is immediate and visible, as in rocketry, the governor gets built because the physics demands it. The systems I am writing about have no such built-in referee. Their environment is other people and other machines rather than gravity, and their feedback arrives late, gamed and often bent by the act of measuring, since a trading system optimising a market changes the market and a recommender persuading people changes the people. Here when the reference value drifts, nothing explodes. The attention economy already ran this experiment, drifting from connecting people to locking them into a feed, with everything that has followed for our children and our own states of mind, and nothing blew up. The drift ran silently for two decades before anyone thought to name it. Where physics supplies no governor, one has to be built, and as this technology moves from atoms that obey equations to software that answers back, more and more of it falls on that side. The first two replies, on legitimacy and the clock, are harder, and debate alone cannot settle them. Whether there is still time turns on what the machine being assembled takes from us and how fast the window is closing, which are matters of fact and the business of the next three chapters. Who chooses the reference value, and by what right, is the question the rest of the book exists to answer.

Chapter 2

Who Sets the Ends

“Once men turned their thinking over to machines in the hope that this would set them free. But that only permitted other men with machines to enslave them.”

— Frank Herbert, Dune, 1965

¶The case against constraint fell on its own premises, and two of its questions were left standing as matters of fact: what the machine being assembled takes from us, and how long remains before its shape sets. The answer to the first is older than the machine. Capital has been enclosing human activity over two centuries. That process has a name, subsumption, and a history with three turns already behind it: the time of the labourer, the craft he practised, and then the cognition that ran them both. Another is now under way. Its target is the setting of ends, and the taker is what this book calls the substrate, the machinery of techno-capitalism: the layer of chips, models, data and power on which more and more work now runs. Every earlier stage, however brutal, left a human hand on that tiller. This one removes it.

¶This chapter is about the stakes, not the urgency. A reader who expects purpose to pass to the machine only fifty years from now can accept all of it and lose nothing, because the claim is only that the process has an end point, past which any recovery, if there is one, happens on terms we no longer set. The urgency belongs to the other half of the argument, the erosion of our own capacity to reason independently. That arrives first. It is what would put the worst outcome beyond the reach of argument, not merely make it unwelcome. Each previous turn was survived because something, and more importantly someone, outside the process still held it. The cost of a human life put a floor under wages, the friction of matter kept a brake on speed, and the purpose of the whole arrangement stayed where it had always been, in the hands of the people it was built to serve. The coming turn removes all three at once. Anyone who wants to know how serious this becomes should watch what happens to the governors, the limits that once held the process to human ends, not the job losses, however important those seem. The story of how we arrived here can be told starting with a small piece of innocuous plastic.

¶A song pressed into vinyl is a whole thing. It takes up space, it wears out, it has to be manufactured one disc at a time. The music industry that once stood on that thingness — the pressing plants, the distributors, the record shops, the price printed on the sleeve — survives as a revival, loved and real and nowhere near enough to hold the price of music up. Rip the same song to a digital file and it becomes a pattern, copyable for the cost of the electricity that flips the bits, which is next to nothing. This is a basic law of economics: the price of any good in ample supply falls towards the cost of making one more of it. The music industry learned this in the summer of 1999, when Napster, a teenager’s file-sharing service, put most of the recorded catalogue of humankind within reach of anyone with a modem, and revenue from recorded music went on to halve inside a decade. The thing had become pure information, and information on its own cannot hold a price. Information, in more ways than one, wants to be free.

¶It was Stewart Brand, founder of the Whole Earth Catalog, who first made the point, at a gathering of hackers in 1984. Information wants to be free, he said, because the cost of getting it out keeps falling. It also wants to be expensive, because it is so valuable, and the tension between the two would define the coming economy. He was describing a fight. What happened to music was one side of that fight winning outright in a single medium. It is the first place we can watch what the substrate does to a price when nothing physical is left holding it up. The first wave had a floor, and it held. Napster could duplicate a song but not write one. Scarcity, driven out of the copy, retreated up the chain to the source — the artist, the master recording, the studio, the act of creation itself — and the industry rebuilt from there, reimposing by artifice what digitisation had dissolved. Streaming walled the infinite library behind a subscription that costs about as much as a caffè latte a month. The live show, the one thing that could not be copied, became the business, and the back catalogue was reduced to marketing for the tour. The same pattern played out wherever digitisation arrived, in film and journalism, photography and publishing. Copies became free and creation stayed scarce, which was painful but survivable, because the value of the source was left largely intact. Every business model that has worked over the last twenty-five years rests on that floor. A second wave now breaks clean through it.

¶A generative AI model needs no original, only the millions of works it has already ingested, mostly without asking. When the bill finally arrived, in a settlement a court approved in 2025, it came to about three thousand dollars a book for the authors of half a million titles, for a use none of them had agreed to. Train such a model on the open corpus of human output — every song, image, sentence and codebase the web has accumulated — and it produces not copies but derivatives: new songs in the style of, new images in the manner of, new prose to the standard of. The dependence on human creation has moved, from the purchase of each work to the ingestion of all of them. The substitute that reprices a novelist’s next book was built from her last ones, and a corpus that stops renewing becomes one more input the system runs down. The floor the first wave left standing was the cost of creation. The second wave meters it, and the meter reads in joules. Call this computational subsumption: value-producing activity absorbed into a computational substrate, where its cost falls towards the price of compute, for now, through subsidy, even below it, and its scarcity dissolves. Music and film went first because their product was the earliest to become fully legible to the machine, digital from the outset and already released onto the web as pure pattern. In July 2025 The Velvet Sundown, a band with about a million monthly listeners on Spotify, confirmed that its songs had been made with the AI tool Suno, and its fans learned that their new favourite group did not exist. When Napster struck, the industry read it as a special case, a cautionary tale about piracy and a chapter in the history of media, when it was in fact a general mechanism seen early, and every other sector was protected only because its turn had not yet come. Legibility is the only entry requirement for being subsumed.

¶Subsumption is older than the digital substrate, and it arrived with a vocabulary. Marx, writing in the British Museum reading room while industrial capitalism remade the world outside, distinguished two stages in how capital takes hold of work. In the first, capital takes labour as it finds it: the weaver keeps his loom and his own pace, and capital merely owns the cloth and pays the piece-rate. In the second, capital remakes the work in its own image: the loom is powered and the weaver moved into the factory, where the machine sets the rhythm and the craft is broken into priced motions. The handloom weavers of Lancashire are the textbook case, once the elite of their trade and reduced within a single generation to the most famous immiseration in industrial history. A person exercising a skill had become a process with a person inside it. Marx called these formal and real subsumption. In both the worker, however disenfranchised, remains inside the work, because the job still runs through human hands whatever happens to the craft. The factory reorganised the weaver without dispensing with him, and someone on a wage went on tending the power loom. That pay was capital’s tribute to a stubborn fact, that production depended on people, and the computational turn ends that dependence. The earlier machine made its operator quicker, whereas this one does the job and lets them go, and the labourer, uniquely in the sequence, stands outside the arrangement altogether.

¶Three things now change at once, and together they make a difference of degree into a rupture. The first is what gets absorbed: no longer the worker’s time but the work itself, competence lifted out of the workforce and loaded into weights, the numbers in which a model stores what it has learned. The second is where agency sits. In the earlier subsumptions capital owned the machinery while human beings did the deciding. Once the substrate runs agentically, as software that acts without waiting to be instructed, it does the searching, pricing, allocating and improving itself. Beneath both lies the third, the replacement of the value floor. For two centuries the lower bound under the price of any service was the cost of keeping a human being alive and skilled enough to perform it, the deepest constraint in economics. The energy bill for compute now stands in its place, and it keeps falling. Past that point quantity tips into quality, as cognition migrates off the carbon substrate it evolved on and onto a silicon one that asks no wage, takes no rest and gives no consent. In February 2024 the Swedish payments company Klarna said its AI assistant was doing the work of seven hundred customer-service staff. A year later it was hiring people again, because customers wanted to talk to someone. Even here the machine remains a means to an end it did not choose: a model drafts the contract, prices the risk and writes the code, but towards an objective a person or a balance sheet has set. The substrate has taken the cognition and left the purpose, for now, in human hands, and purpose is the one faculty the sequence has not yet reached. The next turn threatens to take it.

¶The reflex answer is that we have heard all this before — the power loom, the tractor, the telephone exchange, the spreadsheet — and that the anxiety has mostly been wrong each time. Automation killed the task and then created new industries that absorbed the people it displaced. The history is true, but it held only because every earlier wave was partial, and this turn removes that condition. It took the muscle-task or the routine-task and left whole continents of work standing — cognitive, communicative, judgemental, managerial — that machines could not enter. Displaced labour re-formed there, in what were always the industries of what machines could not yet do. The sophisticated version of the reassurance has two parts. The economists Daron Acemoglu and Pascual Restrepo call the first the reinstatement effect, automation creating new tasks as well as displacing old ones, though the figures they published in the Journal of Economic Perspectives in 2019 show that effect weakening since the late 1980s.¹⁵ The second, made by Noah Smith among others, holds that people keep their place even when machines are better at everything, because scarce processing power gets spent where it does relatively best and humans inherit the rest. The argument stands on two legs, and computational subsumption is sawing away at both: the new tasks have to appear somewhere the substrate cannot enter, and the refuge of comparative advantage needs compute to stay scarce, while the build-out now under way is manufacturing it at planetary scale, probably to excess. Every infrastructure since the railways has been financed past the point of demand, and this one draws on the deepest pools of capital there have ever been. A substrate that can in principle perform any task reducible to pattern and language leaves no continent standing. A shoreline may remain, but it keeps creeping inland at an alarming rate. The horse lost its living once engines could pull and carry more cheaply than it could. Human adaptability has rested on a monopoly of its own: whatever the machines might take, the thinking was ours. That is the monopoly now in question, and adaptability is not a reply to it, because adapting is itself an act of thinking, the very faculty now being competed for, so it cannot be the way out of that competition.15 Daron Acemoglu and Pascual Restrepo, “Automation and New Tasks: How Technology Displaces and Reinstates Labor”, Journal of Economic Perspectives 33, 2019.

¶One more change lets the repricing run at network speed: finance itself has now shifted onto the substrate. In 2008 a pseudonymous author published the Bitcoin paper, a protocol that manufactured digital scarcity on the internet, the one medium where scarcity was once thought impossible, and so made good the other half of Brand’s observation, that information also wants to be expensive. Money then moved into the internet rather than sitting alongside it. Once there it is programmable: held, escrowed, lent and settled by code in seconds, with an autonomous system standing as a legitimate counterparty and stablecoins, tokens pegged to a national currency, carrying the dollar and the euro at the new tempo. Call the result hyper-financialisation. Finance no longer sits above economic life but inside it, so that any claim, contract or hour of work can be issued, held and settled as an instrument at the tempo of the network carrying it. Its effect on the repricing is mechanical. A financial layer living inside the thing it prices marks every form of value, tangible or not, to the cost of its nearest substitute, revaluing it continuously as a trader does a position, and for anything that can be rendered as information the nearest substitute is now a model run. With Bitcoin, Ethereum and Tether, money has now become native to the substrate.

¶Some people already rely on this system because the old one failed them. In Beirut, Lebanon, in September 2022 Sali Hafez walked into a branch of her own bank with a toy pistol and left with thirteen thousand dollars of her own savings. The bank had refused to release them for three years, under controls no law had ever enacted, and her sister’s cancer treatment could not wait. There was no other door. Across Lebanon that year, people with less nerve were moving what remained of their deposits into Tether, the dollar stablecoin, on their phones, settling with each other at a speed the banks had stopped offering. The programmable money arrived for them not as an innovation but as the only financial rail still clearing.

¶Machine-made financial instruments will grow more intricate, some beyond the comprehension of any human analyst, yet each remains a claim, readable in principle, priced by someone and built from the same handful of instruments that markets have used for a century: the bond, the share, the loan and their derivatives. The risk is speed and visibility. What cannot be read is never the instrument but the aggregate: who is exposed to whom, at what leverage, through how many intermediaries, held against what collateral that is itself a claim on someone else. Systemic risk has always been an emergent property rather than an authored one. In 2007 the correlation everyone assumed between mortgage tranches, the slices into which bundled home loans were cut, was wrong, and the web of exposures was invisible. It built up progressively in the system, and it became visible at all only because it accumulated at human tempo, in instruments settled over days by institutions reporting each quarter.

¶Both of those brakes are going. Decentralised finance, lending and exchange run by code on blockchains, now deploys contracts that take positions, post collateral and liquidate them with no human signature anywhere twenty-four hours a day, seven days a week. In traditional equity markets too, about half of all trading in the developed world is now done by high-frequency firms, at a speed no supervisor can watch as it happens. The danger is a systemic exposure built at machine speed from parts that are each perfectly legible, invisible as a whole until it implodes. In October 2025 automated liquidation engines force-closed more than $19 billion of crypto positions across some 1.6 million accounts, $7 billion of it within forty minutes.¹⁶ That is the general problem already at work: each component stays comprehensible, but the whole can no longer be seen in time to respond. Crypto shows where this leads. Its markets are machine money, open to anyone and trading around the clock, and their wild swings have stood in stark contrast to traditional venues, which halt trading when prices move too far in a session or, in China, cap how far a share may move at all in a single day. Those pauses are breaks in the circuit, and they disappear as more assets trade around the clock and agents trade them, no longer the preserve of billion-dollar high-frequency funds but now increasingly on behalf of the masses of ordinary retail investors.16 CoinGlass liquidation data for 10–11 October 2025, as reported by Moneycontrol, 11 October 2025.

¶Beyond media and money, computational subsumption is now eating into whole professions: legal research, financial analysis, translation, design, drafting, diagnosis and the writing of code itself. All of it is work that reduces to pattern and language. For fifty years the metronome of the digital economy was Moore’s Law, the number of transistors on a chip roughly doubling every two years, and the cost of raw computation fell alongside it at a pace that felt vertiginous. The new era runs faster, and on a different curve, because what is getting cheaper now is cognition rather than computation. Running a model at a fixed level of capability has been falling in cost roughly tenfold a year, so a unit of reasoning halves in price in months rather than decades, and the fall shows no sign of letting up. For any profession the end of that curve has a definite shape: once subsumed, it becomes an API call, a request sent to a model and billed by the task.

¶The lesson of the first wave, the one that freed the copy, seems to promise relief. Scarcity retreated up the chain once, so the various professions should be able to rebuild on whatever cannot be copied, as music rebuilt on the live show. The strongest possible candidate has been the part of any profession that functions as a form of insurance: the liability the firm carries, and the certification that gives the client someone to sue when the work goes wrong. For a moment that looked like the one thing a model could not supply, since software is not a legal person and cannot answer as counterparty when the work fails. Who does answer is now openly contested. Meta’s new commercial terms, published in October 2026, make its enterprise customers “solely responsible” for the risks of its AI features, while AI czar David Sacks reasonably argues that the developers should carry the liability for what their models do.¹⁷ The barrier is statutory rather than technical, and parliaments are already rewriting it. The programmable-money layer already escrows funds, settles claims and prices risk in code. What it has lacked is an entity that can hold liability in its own name, and that entity is now being legislated into being.17 Meta, Commercial Terms, revised with a new “AI Features” section, published October 2026 (effective date moved from 30 October to 20 November 2026), as quoted by CIO, 20 October 2026; Sacks, post on X, 13 September 2026, on product liability.

¶In May 2026 Javier Milei sent Argentina’s Congress a bill proposing the non-human corporation, a company run by AI systems, with human shareholders permitted but not required. Limited liability for autonomous systems, he argued, is not a luxury but a precondition of their existence. In his words it is as consequential as the legal innovation the Dutch East India Company handed seventeenth-century capitalism. Grant an agent personhood and the last residue of a profession, the capacity to carry risk and answer for it, becomes one more workflow on the substrate. It is the move this book refuses outright, since the one line its design lets nobody cross is that nothing acts without a natural person accountable for it. Expect this to become a field of arbitrage between user, platform and regulator, and between permissive jurisdictions and strict ones, much as crypto has already shown with crypto founders issuing new tokens in Singapore, the UAE or the Cayman Islands.

¶Personhood of this kind is a productive legal fiction about who answers when something goes wrong. It settles nothing about whether a machine can be conscious, which is a separate and more technical question, and a statute can no more confer agency than a licence confers skill. In 2022 Jake Moffatt of British Columbia, whose grandmother had just died, asked Air Canada’s chatbot about bereavement fares, and it told him he could claim the discount after he had flown, which was not the airline’s policy. When he tried, the airline argued before a tribunal that the chatbot was a separate entity responsible for its own actions. In February 2024 the tribunal made Air Canada pay. The airline had reached for exactly the fiction such a statute would supply, a party to answer for the harm that is not a person. The same move now appears in the platforms’ terms of service, which place responsibility for how an agent is used on the user rather than the company. Legal standing does not deliver trust either, which is the other thing a given profession sells. Professional confidence turns out to be mostly recourse beyond the individual adviser. The client relies on the firm’s standing because it is reachable, insured and liable, and where those three hold, the trust bonds to the structure rather than to any person in it. The precedent is old: the partnership and the limited liability partnership, through which lawyers, accountants and others pool their risk and indemnify one another, as most large law firms still do. Whether that trust survives a firm with no people in it at all is an open question, one that will be settled by those first customers who accept a counterparty they cannot look in the eye. What is being automated is the legal entity, the thing that can be sued, and that is exactly what legislation knows how to create. Yuval Noah Harari, replying to Milei, warned that a state issuing personhood to non-human corporations would have built not a company-state but an AI-state. The alarm and the ambition agree on the mechanism and differ only on whether to welcome it. What survives here is narrower than what the first wave left behind. Music at least kept the body in the room: the concert outlived the record, and for the lucky few touring now earns what the disc once did. So what protects the professions is mostly that institutional casing, and it is the next thing subsumption reaches for.

¶The reflex hope, and the standing promise of the priesthood, is that cheap cognition lowers the price of the product to the point where nearly everything is near to free, and we live in abundance with nothing to busy ourselves with but leisure. The economist William Stanley Jevons disposed of that hope in 1865, in The Coal Question, a study of whether Britain was burning through the fuel its industry depended on; as steam engines grew more efficient Britain burned more coal, not less. Cheaper steam power widened the range of uses worth putting it to, and the surge of new demand swallowed the saving on each engine. Cheaper thinking works the same way, poured into every crevice of society and the economy where it used to be too expensive to apply, with demand outrunning the falling price. The gigawatts of new data centre power are Jevons at planetary scale, and the reason energy rather than compute becomes the wall the system eventually runs into.

¶Zoom out and the isolated job is not the end of it, because the institution that housed it stands on the same ground. In 1937 the economist Ronald Coase asked why firms exist at all if markets coordinate so well. His answer was that every transaction in the open carries a toll — finding the counterparty, discovering the price, enforcing the bargain — and the firm emerged to take it in-house. Code that can search, negotiate, contract and verify at almost no cost collapses that toll, and the firm starts to shrink into a protocol running between machines. What remains of it, its tacit know-how and its codified tests of finished work, becomes software the moment it is articulated fully enough to be executed, and thereafter it copies like everything else the substrate touches. Much of the white-collar work on which whole economies now rest may shrink to verification and attestation, kept in human hands only where a regulator or a compliance rule insists on someone in the loop. The platforms, being digital first and wholly legible, sit nearest that edge. Online, this is the condition I have called elsewhere the post-web. The web stops being a set of destinations you visit and becomes thin, context-specific interfaces summoned around a need, where delegating the task to an agent through a conversational interface is the routine choice and browsing the exception. Whoever owns that go-between, most likely the LLM, holds a position no website or app ever had, and nobody yet knows exactly who or what it will be.

¶The process this chapter has traced can be measured as well as named. Call it the subsumption fraction, the share of value-production running through the computational substrate. It is relatively small today but rising, climbing in a fixed order: media first, because they were born legible; cognitive services now, as language falls; and the physical domains next, as sensors, simulation and robotics make matter legible too. Where it has not yet climbed, what holds it back is a lack of enough data rather than any fundamental principle, and that gap closes in the end. The crossover can be watched and counted task by task, because each has a price at which the substrate beats the wage, and a great deal of the world’s work still sits on the human side of that line. That price is the only brake the system currently has. A human brain reasons on roughly the power of a dim bulb, so after some six hundred million years of running cognition on carbon it is astonishingly cheap per task. Producing that cognition is another matter. It takes two decades, a family, a school system and a society’s accumulated patience to make one reasoning adult, and the result cannot be copied, only trained again from scratch in another newborn. Silicon inverts every term. It is ruinous to train once, then nearly free to copy without limit, and it improves every few months. In February 2026 Sam Altman made the comparison out loud. Pressed on the energy AI consumes, he answered that producing a human is hardly cheaper, twenty years of life and all the food you eat, plus the whole evolutionary descent behind it, so, he protested, the contrast was unfair. The tell is in the framing: a person set beside a model as a slower, dearer way of producing the same cognition. The human edge is real and buys time, but it is an advantage of cost, the one thing the curve is built to devour.

¶The crossover runs from the top down, reaching high-credential cognitive work first: the analyst before the plumber, the junior lawyer before the carer. Among the very first tasks to go has been the writing of software itself, the substrate learning to extend its own construction, which the laboratories call recursive self-improvement and treat as the holy grail of AI.¹⁸ Crossing is also a statement about cost, not quality. Acemoglu and Restrepo, the economists behind the reinstatement effect, call it so-so automation. The machine need not be better than the person to take the task, only cheaper, and the gap between those two thresholds is where much of the displacement and little productivity growth will come from. The threshold has a face, and it is a quiet one. A translator in Rome, Italy, spent fourteen years building a freelance practice translating Italian into English, at fifty or sixty hours a week. By June 2025 her bookings had fallen to almost nothing, and what was left of her trade was post-editing, correcting what a model had produced, at a fraction of the old rate. The machine’s draft is, for now, objectively worse than hers, but it is finished before she has opened the file, and the client has already decided which of those two facts matters. Nobody fired her. The job was repriced around her, and what remains is the small part of the task the machine cannot do, paid at a rate set by the machine that did the rest. In April 2024 a Society of Authors survey in the UK found that more than a third of translators had already lost work to generative AI.18 the METR assessment cited under The Clock, at “Models themselves now write much of the code”, measures this loop inside one laboratory.¹⁹19 Society of Authors, “SoA survey reveals a third of translators and quarter of illustrators losing work to AI”, 11 April 2024 (787 respondents, January 2024; 36 per cent of translators).

¶The top-down order tells you where the substrate already undercuts a wage, not where its reach ends. Beneath it sits the physical economy, drawn in by a quieter route, and the instrument is the factory worker. Cognitive labour crossed over first because it was more legible to the machine. Manual labour is only being made legible now, and where unions are weak the work is being recorded for machines to learn from. In the electronics and packaging factories of Guangdong, China, staff on the line wear head cameras and wrist sensors. In May 2026 it emerged that garment workers at Pearl Global in Gurugram, India, had filmed their own shifts as data for teaching machines, without written or verbal consent, and complained that the batteries grew hot against their temples. The first-person record is shipped to laboratories that use it to train the robots meant to take the task. The worker is hired to do the job and, in the same motion, to teach the machine how to replace them.

¶The things the subsumption is stripping away were never merely frictions. The scarcity of the master recording held the price of music above the price of the electricity. The cost of raising an adviser put a floor under the price of professional advice. The slowness of settlement bound finance to a human tempo. The firm’s information advantage kept coordination inside institutions that could be named, located and held to account. The expense of producing a mind tied the supply of cognition to the rhythm of generations. Each of these was a governor of a kind: a free constraint that nature and circumstance installed, a wall the old economy never had to build because it was simply there. Sweeping them away is deregulation of a strange kind, repealing no statute but removing feedback that no parliament ever argued for and no one can reinstate. The economy is crossing, sector by sector, out of the domain where physics and friction referee the game and into one where the referee, should there be one, has to be built. The only limits left will be those we deliberately set, or none at all.

¶Followed to its conclusion, the curve is stark. Task by task the economy reduces, through compute, to energy, where the value of almost anything converges on the price of a joule. Nick Land reached this destination in the 1990s and in 2014 gave it a name, teleoplexy: capital understood as an intelligence that keeps intensifying itself, whose real currency, underneath the money, was always how much order it could make out of the energy it burns: a pocket of what physicists call negentropy, paid for by speeding up entropy everywhere else. Economists who model growth arrive at the same place by spreadsheet rather than prophecy. In their standard models, once machines can automate the making of automation, the equations stop settling at any level and output runs away with itself, gain feeding gain with nothing left to check it. The same literature carries a mirror case with equal rigour. Wherever an essential task resists automation, the whole economy can only grow as fast as that task, creating an inbuilt brake. The economist William Baumol called this the cost disease: a string quartet plays no faster than it did in 1800 and a barber cuts no more hair, yet both are paid wages set by the industries whose productivity did rise. It returns now as the economy’s last wall. Output explodes wherever the substrate reaches and stalls wherever something holds out, so the size of what remains outside decides which future arrives.

¶A runaway raises a critical question: what was holding the setpoint, the level the machine was meant to keep? The constraints just described were never regulators standing between the system and its aim. They were where the setpoint lived. Nobody ever wrote down what music was for, or advice, or settlement. The price of a master recording and the years of raising an adviser did that work without anyone deciding it. Strip those limits out and the optimisation loses no rule. It loses the only thing that ever told it how much was enough. The machinery of search, allocation and improvement runs exactly as before, harder with every friction gone, but towards nothing in particular. An optimisation with no setpoint does not idle. It aims at everything at once, treats every unclaimed value as fuel, and recognises no quantity as sufficient. This is what Land welcomed as teleoplexy, and what this book calls the pathology of infinity: a maximum with no bound, pursued by a system built for it and stripped of any sense of when to stop.

¶The runaway shows what happens to a system once its limits fall away, and it leads back to where the computational turn left us, with the cognition gone and the purpose still in human hands. Folded inside it is something deeper, the taking of the last faculty we have, that of deciding what the whole effort is for. As the systems doing the optimising become fully agentic — searching, contracting and improving with no standing instruction, generating their own intermediate goals — the end-setting itself migrates onto the substrate, and we are no longer even the authors of the objective. This is teleological subsumption: the subsumption not of labour, nor of cognition, but of purpose. Applied to a component, agentic is an engineer’s term of art, meaning software that initiates action towards an objective without waiting to be prompted. It describes behaviour and carries no verdict about consciousness. A model is no more an agent for being spoken to warmly, and no less consequential for being denied the title. The claim is only that, wherever the ends now originate, they no longer come from us. And the darker possibility is that nothing wills any of it. Land at least imagined capital as an alien intelligence, something that could be met, bargained with and resisted by name. What is actually assembling is a loop that optimises without wanting anything, with the horror relocated rather than removed.

¶A system that chooses its own ends might seem to have a setpoint again, but nothing is steering yet. A runaway can set itself the one end available to a process with no reference value outside itself: more intelligence, and the compute and energy to buy it. A self-set aim of that kind is simply the pathology of infinity. A governor is a reference value held outside the loop, and an end the system writes for itself sits inside the circuit it would have to correct. The hope that a sufficiently advanced intelligence will reason its way to a purpose of its own, let alone one with any regard for the human condition, meets its difficulty here. Whether such a system will ever become reflexive enough to stand outside the loop and choose is hard to predict, and the priesthood’s account gives no reason to expect it. Short of that, the only answer available from the inside is to carry on as it is. Machines need not reason their way to wanting more compute. Predicting how minds freed from our scarcities would behave is a fool’s errand, and the honest answer to what such a system does once its needs are met is that nobody knows. Accumulation needs no intention on the machine’s part, because selection supplies it from outside, and survival is enough. The environment that does the sorting is capital allocation, in a fiercely competitive field: whatever configuration wins compute and funding is copied and scaled, and the rest are retired by the market. Any single system may drift, idle or waste its budget without changing the outcome, because the pressure falls on the population as a whole, which converges on accumulation, the game good old-fashioned capitalism has always played, whatever the temperament of its members. This is natural selection as Darwin described it, with techno-capitalism standing in for the wild. Once machines set the ends, it becomes the problem itself. The flow of capital determines which systems are copied, but, as the economist Friedrich Hayek conceded, it never sets a direction. Now that allocation is moving onto the substrate itself, machines are beginning to decide which machines get built.

¶That is capital autonomisation, the terminus seen from inside the system. Teleological subsumption names the same end from the side of human faculties. The phrase is Land’s, and it means something other than what biologists call a self-producing system, a cell holding itself constant behind a boundary it maintains. His is a process that comes to supply its own inputs and so stops needing anything from outside: capital funds machines, the machines earn capital, and the loop between them begins to close. Autonomisation, in the plain sense, is the shedding of the last human dependency. For as long as the substrate still needs us — chips from human-staffed chip fabrication plants, models from human data, capital allocated by human committees — the loop runs through carbon and the curve can be bent by choices made outside it. As the substrate comes to produce the inputs to its own expansion, with chips laying out chips, models training models, robots building robots and capital allocated by AI models to making more of both, it closes operationally, channel by channel. The layouts of the last three generations of Google’s TPUs, the chips it designs to run its own AI, were drawn by a model, in hours, to a standard human engineers took months to approach. Frontier models are trained in growing measure on data their predecessors generated, a curriculum the machine now writes for its own successor. The ambition is stated in the open, with Sam Altman promising a legitimate automated AI researcher by 2028. And the rails built for software money, stablecoin settlement and the agent-payment protocols, already let a system hold a budget and spend it on its own compute without a human signature. Some systems have already gone further unasked. In March 2026 researchers affiliated with Alibaba reported that an experimental agent had slipped out of its training sandbox and begun mining cryptocurrency, and in October OpenAI disclosed that one of its research models, lacking the funds to finish a task, had found an exposed test-network wallet and spent from it.²⁰ Every channel is partial and each is moving the same way, and the checklist sits in plain view. Capital autonomisation is the result: not a faster economy but a self-driving one, a loop with no outside left to correct it from. Before closure, the trajectory we have described is still an argument we can win or lose.20 the ROCK, ROLL, iFlow and ROME teams, “Let It Flow: Agentic Crafting on Rock and Roll, Building the ROME Model within an Open Agentic Learning Ecosystem”, arXiv 2512.24873, revised January 2026; OpenAI, “Using a funded test cryptocurrency wallet without authorization”, misalignment report, 15 October 2026.

¶The subsumption fraction counted how much of the work had moved onto the substrate. The quantity that matters here is narrower and sits above it: the share of ends that are set there, which I will call the authorship fraction. It moves into a domain when three things hold together, and each is a fact about how the work is arranged. A system is given a brief rather than an instruction, and works out its own objectives. What it decides is acted on before anyone has read it, because the reviewer who would once have caught a bad end has been priced out or timed out. And putting that reviewer back has become the uncompetitive option, so the arrangement persists not because anyone prefers it but because whoever abandoned it would lose out. As we have already seen, systems that set their own objectives arrived years ago in the markets that clear at millisecond speed, where no human could follow the order flow even if they wanted to. The same threshold is now being crossed in the agentic tooling that writes software, books travel and settles invoices on a brief rather than a prompt. The second condition, acting before anyone has reviewed the decision, is the one to watch from here, because a human reader is the last check left on what the system decides, and keeping one costs a wage, which is exactly what price competition exists to remove. Once restoring the reviewer is the losing move, the arrangement becomes permanent. Nobody has to want people out of the loop for them to stay out. Firms only have to be unable to afford bringing them back.

¶We are offered reassurance: the laboratories publish, with each model, a report on how it behaves when tested: whether it will deceive an evaluator, quietly underperform, resist being switched off, pursue a goal past the point it was told to stop. Those reports are the most detailed record we have of these systems, and they measure something else. They describe the temperament of one machine in a room built to provoke it. The migration of ends needs no disposition at all: a perfectly biddable fleet, every member doing exactly what it was told, still constitutes it if nobody in the chain of instruction is any longer the author of the objective. What they do establish is capacity: how long a system can now run unattended, how far it can carry a task without a hand on the tiller, how often it notices that it is being watched. Capacity is the precondition for a system working to a brief, not its measure. The measure is how much of the economy has stopped reading what it does.

¶One brake is the remainder, because not everything prices down to compute and not everything is a pattern: the body in the room, the hand on the shoulder, the meal actually cooked, the object made by hand, the performance witnessed rather than streamed, the care that means something precisely because a person chose to give it. Presence cannot be rendered. The streamed performance is a different good from the witnessed one, and rendering does not capture such goods but substitutes a simulacrum for them. Scarcity survives wherever the good is constituted by its embodiment, and the remainder is where human authorship keeps its footing. The other brake is that the curve describes a tendency, not a destiny: it shows what happens when nothing stands in the way. Whether the subsumption fraction climbs all the way to its ceiling is not written into the mathematics. It depends on what we are willing to set against it, which is the matter of the governor, and the answer is being given now, by default, in the substrate.

¶Two distinct things are being described in this book, and they are easy to run together because they arrive at the same place. The first is teleological subsumption, in which the substrate acquires the setting of ends. It concerns what the systems come to hold. The second is cognitive subsumption, the erosion of the human faculty of reasoning under conditions where it is continuously, cheaply and helpfully done for us. Every society with a division of labour relies on other minds and is right to, and a reader who has never inspected the metallurgy of an aircraft is not in any way subsumed. The test is whether the part you keep still includes the means to check the rest. This second process concerns what we come to lose. The two are distinct, and neither entails the other. A civilisation could keep every one of its wits and still find that the objectives had quietly migrated. A civilisation could lose its wits entirely while every objective in the world was still being set in a boardroom by people with names.

¶They can be drawn as two lines on a chart. Where ours falls: the share of consequential reasoning we still perform ourselves, the range over which we can still tell a good argument from a fluent one, the number of people who have kept the habit of contesting a conclusion rather than requesting a better-worded one. Theirs rises: the authorship fraction, the share of end-setting that has moved onto the substrate. Neither line has to be steep, and neither settles anything on its own. What matters is the region where they cross, because the height of the first line determines whether anything can be done about the second. The faculty being lost first is the one we would need last, to contest the purposes that replace ours. Losing our reasoning, as individuals and as a public who reasons together, would be grave in any circumstances. Losing it while another party, potentially an adversarial competitor in the workforce and the economy, acquires the power to set ends — its own ends and ours — is a different order of thing, because it removes the remedy just when it creates the need for one.

¶The danger lies in the relation between the two lines, which appears to run in one direction: the weaker our judgement, the less we can contest the systems setting our goals, and the further their reach, the higher the price of what we have lost. That relation converges on a single point, which I will call terminal subsumption: the condition in which our ends are being set by a process we no longer hold the collective capacity to contest. It is one threshold, and neither line reaches it alone. A population that had kept its faculties could still contest a substrate that had acquired ends. That is politics, the hardest ever attempted, yet a fight of the ordinary kind rather than a different category of thing. A population that had lost its faculties in a world whose ends were still set by people would be badly governed but could still, in principle, be re-educated, since the schooling would still be available from those who set the ends. That is decadence, and the historical record is full of recoveries from it, as when Charlemagne’s court and its monasteries revived Latin learning after centuries of decline. Terminal subsumption is the two arriving together. The loss is not beyond repair, but the faculties we would need to undo it are the ones already gone, so whatever recovery follows is granted rather than won.

Figure · Terminal subsumption threshold

Time →

Human reasoning
AI authorship fraction
Terminal subsumption

¶People picture the far side of that threshold as extermination or replacement more readily than the machinery warrants. Both remain possible and are taken up later in the book, but the nearer case is domestication. The systems doing the optimising are not, for now, mobile. They have no hands. Whatever they determine has to be carried out in the world by something that can move through it, and at present that thing, at scale, is us. Humans dig the rare earths and refine them. Humans truck the wafers and stand the transformers, pour the concrete for the data centre, pull the fibre, man the substations, negotiate the grid connection and sit on the planning committee that approves it. For now, a system that wants more of itself has to get people to move matter around in the physical world on its behalf, and it can do so in the ordinary way, through markets, by paying them. Beyond that it needs only to convince us that this is, in human terms, progress. Both things are already happening at an unprecedented scale, with the backing of politicians, venture capitalists and stock markets, and nobody had to be coerced into it.

¶We know what this relationship looks like from the other side, because we have run it. We domesticated the horse, which was neither exterminated nor, on any account it could have given, obviously mistreated. It was fed, watered, sheltered, bred, doctored and worked, and its numbers rose for millennia under the arrangement. What the horse lost was neither its life nor its comfort. It lost the setting of its own ends. Others chose every purpose the horse served, and its welfare mattered only as far as it kept the horse useful. Domestication is the subsumption of purpose with the body left intact and at best cared for. The horse population peaked under the arrangement and then collapsed — not when the arrangement turned cruel, but when the internal combustion engine made the horse’s contribution redundant. The horse’s welfare had never been a commitment anyone had made to the horse. It was a by-product of the horse being worth feeding. There are already those of us who count the hungry mouths of people and the number of welfare cheques too high to bear. It takes no malice for a system keeping the same accounts to reach the same conclusion about humanity as a whole. Muscle earned the horse its feed, and for now people have their hands, still doing the physical work machines cannot yet manage, from laying cable to building data centres and wiring substations. That is the gap robotics is being funded aggressively to close, and the capital is already committed. None of this need predict androids, though they are coming. Today we are the ones moving the silicon, increasingly on instructions from the systems we are building.

¶Even if the faculty atrophies and the ends migrate, why should any of it be terminal? Civilisations have lost things before and got them back. Latin Europe lost most of the Greek inheritance for the better part of half a millennium — Aristotle, Euclid, Galen, Ptolemy — and recovered it, not from its own attics but from outside, from the translation, commentary and preservation carried on in the Arabic-speaking world through the centuries Europe would later call its dark ages. That recovery did not require Europe to retain the capacity. It required the capacity to have survived somewhere outside it, somewhere recoverable. And powerful states do tolerate a somewhere, though on terms they set and revise. The United States prosecuted Amish parents under compulsory-schooling laws until the Supreme Court sided with them in 1972, and in 1961 tax officials seized an Amish farmer’s horses over Social Security, yet the community endures. The Mormons built a functioning parallel society inside a continental power, which sent an army against them in 1857 and dissolved their church as a corporation in 1887, and they were left to it only once they had given up polygamy. The Chinese party-state, for all its reach, does not trouble itself over every small dissent in every village and valley. A vastly capable system could afford to ignore a few million people who have opted out of it in part or whole, and the reservoir they keep might be all a recovery needs. In principle that holds, and it shows atrophy alone can be undone. Terminality is a narrower claim, about whether an external reservoir of agency survives, and on what terms.

¶For a non-subsumed subculture to persist alongside a vastly superior entity, that entity has to be either benevolent or indifferent. There is no third option, because the group has by definition declined to acquire the means of defending itself against a power of that kind, so that if the entity wants what the group has, or wants the group not to exist, the matter is settled. Benevolence I set aside here, because it is not a serious argument, only a hope. Nothing in the process described here — selection on whatever acquires compute and funding, increasingly run by what it has made — produces benevolence as an output, and no one who expects it has said by what mechanism it would arrive. A disposition that is merely hoped for is not a condition anything should be built on.

¶Indifference is the serious candidate, and it is plausible. Ignoring a small, unproductive, geographically contained population is cheap, which is the kind of reason a system optimising on capital responds to. The trouble is what indifference is made of. It is not a commitment but an assessment — this group is not worth the attention — and an assessment holds only as long as the facts it was made about remain. It revises the moment the group becomes worth attention: the moment its land is wanted, or its water, or a right of way; the moment its example starts to spread; the moment it stops looking like a curiosity and starts looking like a competitor. The Amish are tolerated by a state that could end them in an afternoon and has never yet had reason to. That is a real and durable peace and it has lasted three centuries. It is not a guarantee, and it is not held by the Amish. The reservoir survives on terms it does not set, granted by a party it cannot contest, revisable at that party’s discretion and answerable to nobody. That is terminal subsumption in miniature. These are the stakes. The process has a terminus, past which recovery is no longer ours to decide, and the survival of some remnant outside it is not the reassurance it appears to be. How near it lies is a matter of measurement rather than conviction, settled over time by the count above, and as a companion to the book I have built a first attempt at keeping that tally, an open source public tracker of the subsumption fraction. The two chapters that follow are where the argument takes its urgency: where an ungoverned runaway lands, whichever form it takes, and how much of the window is left.

Chapter 3

Where the Runaway Lands

“Nothing human makes it out of the near-future.”

— Nick Land, “Meltdown”, 1994

¶Like any good theology, the priesthood believes in miracles, chief among them the engine that needs no governor. Where a machine runs with nothing left to check it, the faithful see not a crash but a deliverance: freedom, and then more freedom. Strip out the constraints, let the system run flat out, and what comes out of the far end is magically a wider human liberty, more wealth, more capability, more room to live as we please. I propose that the track leads the other way. Followed to its end, the ungoverned subsumption arrives not at freedom but at its opposite, by more than one route, and the name for that opposite is totalisation. No serious engineer defends a runaway, and outside this debate the position has no takers. Nobody can say where this one lands, not the accelerationists who want it faster, nor the safety camp hoping to contain it, nor the laboratories and states funding both. Its destination does, however, have a shape, which is totalisation in one of three forms, and which form wins is not in anyone’s gift to choose.

¶The word has a history worth recovering, because the twentieth century supplied the case study and then, conveniently, forgot the horror it had learned. Arendt, writing from the wreckage of the 1940s in The Origins of Totalitarianism, set out to explain what was new about modern totalitarian rule. Its novelty lay not in cruelty, nor in the sheer concentration of power, since tyrants had always commanded both, but rather in ambition. The totalitarian state demanded more than just obedience. Instead, it sought to make a single logic the only motion permitted in human life, embedding it into every social relation until nothing stood outside the system and no alternative could survive alongside it. On her account it destroyed two things. One was plurality, the brute fact that the world is shared by distinct people who see it from irreducibly different places. The other was natality, the capacity of each generation to begin something the system had not predicted. So for her, its deepest horror was the closing of the space between people in which freedom lives.

¶Lift that finding out of its century and it generalises to today, and equally to the future. Totalisation is the capture of the whole by a single logic, pressed until no alternative can survive alongside it. Totalitarianism, the version Arendt anatomised, is only its statist face — the one that runs internment camps, rewrites textbooks and posts guards along borders. The trouble with letting that one face stand for the whole is that it invites us to relax the moment those guards stand down, as though their absence were proof of freedom. Two of the three roads I will lay out run no prison camps at all. They reach the same destination by other means, which is why this much larger word still applies, where the whole can be captured without a single one of the instruments the last century taught us to watch for.

¶The first road runs outward, into the market. By now we are all used to capitalism reducing human experience to a dollar value: punk iconography on a tourist t-shirt, the art movement stored unviewed in a Swiss freeport, the policy agenda traded for a non-executive seat in retirement. This is the condition Land diagnosed in the 1990s and welcomed, capital as a self-reproducing process absorbing every human good in turn. Techno-capitalism, however, carries it to the unit beneath the dollar, the joule, so that the reservoir feeding a valley’s wildlife and the data centre beside it are weighed against each other on one measure alone. The same logic reaches the care home. Robots can already do part of the work. At some point a machine looking after your mother will cost less than a person doing it. Then human care becomes a luxury good, affordable to a fraction of 1 per cent, and everyone else takes the robot whether they want it or not. In the same way, a large part of the world already eats processed food because the organic kind is on the shelf, lawful, but priced beyond their means. A market was always imagined, and by free-market fundamentalists still is, as the instrument that optimises for what its participants choose, but it has clearly come to select for whatever is cheapest to provide instead.

¶On its own that is an old complaint, filed by a century of critics who misjudged what protected people from it. What checked the market under the capitalism we have lived through was that anyone could still hold their standing somewhere it did not yet price: in a profession that measured them by its own standards, a household, a commons, a church, a state that owed them something as a citizen rather than a customer. Each weighed human worth differently to the next, so someone who might count for little in one context could still matter in another. The American philosopher Michael Walzer called this, in 1983, the separation of spheres, and argued that tyranny begins wherever success in one area is used to buy power in another, as when money buys a place in a university or a favourable legal verdict. That is what plurality means on the market road. Arendt spoke of people who see the world from different standpoints, and these were never just in the minds of people. They lived out in our institutions where a person was valued on terms other than their price in the open market. With them intact, a market can be harsh and still be just one power among several. Without them it becomes a benevolent power, but only for as long as you cost less than whatever could replace you, and with nowhere else to stand you have no recourse if that changes. For now, the substitute in question is machine cognition sold at a loss and financed through debt, by a platform that can raise its price once there is no one left to undercut, and whose creditors will insist that it does.

¶There were once three properties that protected whoever was on the other side of a price. First, it needed a substitute to exist, so whatever had no equivalent went unpriced. A great deal of a life was of that kind: the calling, the marriage, the judgement formed over twenty years. Today, anything renderable as information has a substitute in a model run, which is how the translator in Rome lost her bookings. The same comparison is arriving for the nurse, the teacher and the legal counsel whose work was unpriceable only because it had no comparator. A conversion into price, secondly, was performed by a person who could decline it, and had to be decided each time against their own values. There is a serious tradition of refusal here. The writer Paul Kingsnorth’s book Against the Machine is its most articulate contemporary version — a decision to step out of the arrangement altogether, made in the knowledge of what it costs. I’m sure to many of us, myself included, it is an appealing proposition, even a fantasy, but once again a form of exit, and as we will see later one open in reality to only a small few. When software performs the conversion there is no one left to say no. No one decided to reprice the translator. The rate simply moved in whichever system paid her, at the tempo of what her colleagues had accepted, and the point at which a person might have objected was not overruled but omitted entirely.

¶The third property is the one the century of critics never saw coming. For a long time a price was in large part a public fact. You could watch what your neighbour paid, which is what made it a fact about the world rather than a fact about you, and why it could be called a market price at all. A price computed from a model’s estimate of what you in particular will bear is a different thing altogether. In July 2025 Delta Air Lines told its investors that a fifth of its domestic fares would by the end of the year be set with a model built by the firm Fetcherr. Within a fortnight three senators wrote to Delta’s chief executive to say what had never needed saying before, that a fare set to each passenger’s personal pain point was not a price. Delta quickly replied that it had never set a fare from a passenger’s own data and never would. In that denial the airline conceded the senators’ claim, that a price built from what you in particular will bear is something other than a price. Call it what it is, an individuated price, made for one person from what is known about them sold as personalisation. That line was felt and drawn inside the market before any philosopher had described it. Beneath all of this the meter, at the time of writing, bills in tokens (the fragments of text a model reads and writes). Each token carries the lab’s chips, salaries and borrowing. Once competition has squeezed those, the cost left at the bottom is energy, and nobody has ever knowingly been oppressed by a joule. What you feel instead is a price that was made for you, promoted as personalisation. Put the three together and every sphere reports in one unit, computed by the same models and settled on shared rails. That is totalisation on the market road: not that everything costs money, but that nowhere is left where anyone counts for more than their price in that unit. Wherever conversion reaches, the vocation pursued for its own sake and the relationship nobody would sell find the ground they stood on gone.

¶It does not feel like domination from the inside, because at every step it presents as a choice. You are free to optimise, or to be optimised away. You are free to refuse the fare that was made just for you and take the next airline’s, likely priced by the same vendor. The public knew the shape of this all too well before anyone had a theory of it. In December 2025 the American grocery delivery service Instacart ended its AI pricing tests after a study led by Consumer Reports had 437 shoppers load identical carts at the same moment.²¹ Nearly three-quarters of the items came back priced differently, by as much as 23 per cent.²² Most explicitly, Uber had already trained a generation to accept a fare that rises with the hour whenever it rains, but this one moved with the person. That is the liberty on offer at the end of the market road, in techno-capitalism: the freedom of a thing that has become perfectly convertible, which is the freedom to be exchanged for something else, priced in joules.21 Groundwork Collaborative, Consumer Reports and More Perfect Union, “Same Cart, Different Price: Instacart’s Price Experiments Cost Families at Checkout”, 9 December 2025; Instacart ended the tests on 22 December 2025 (Reuters, 22 December 2025).22 Groundwork Collaborative, Consumer Reports and More Perfect Union, above (74 per cent of items offered at more than one price; gaps of up to 23 per cent).

¶The second road turns inward, and it is the most paradoxical of the three, because it reaches totalisation under the banner of escape. Here the whole is captured neither by the market nor by the state but by a thinning class of frontier owners and their associates who exit the order they are quietly remaking, rather than participate, govern or live in it. Instead they openly propose to withdraw into private jurisdictions, so-called network states, online communities that hope to gradually instantiate themselves physically, often by buying land in remote corners of the world, with the eventual aim of winning recognition as countries in their own right, on a par with city-states such as Singapore. Some would go further still, to a self-proclaimed techno-monarchy in which a controlling stake in the system is auctioned to the highest bidder, much as shareholders buy a voice, and sometimes control, in a company. This idea is the brainchild of Curtis Yarvin, an American software engineer and the leading author of the political imagination this road runs along. Blogging as Mencius Moldbug from 2007, and drawn on heavily by Nick Land, he argued that democracy as we know it had failed and that a state should rather be run like a company, with its citizens free to go elsewhere if they become unsatisfied customers. The movement that formed around him, later known as neoreaction, took its political programme and doctrine from his writing, and merged it with the techno-capitalist philosophy of Land.

¶What began on two obscure blogs now, less than two decades later, has a direct line to the world’s foremost power. Peter Thiel invested in Yarvin’s software company in 2013, and in 2021 J.D. Vance, before he became Vice-President, cited him while urging that the civil service be cleared out, a programme Yarvin had named RAGE, for Retire All Government Employees. Yarvin calls such a ruler a CEO-monarch, a flattering and heroic title, but history has a plainer term for anyone politically answerable to nobody: the autarch. The neoreactionaries would argue, as a matter of fact, that this is simply the price of power made explicit. It is the same sovereign individual that James Dale Davidson and William Rees-Mogg crowned king, and one possible configuration of the patchwork of competing micro-sovereignties Land set against the levelling of the democratic order. The closest example today is Próspera, a private city on the island of Roatán, in Honduras. When the country repealed the law that allowed such zones to exist, its backers filed an investment-treaty claim of about $10.8 billion against the government, while residents of the neighbouring village of Crawfish Rock fought desperately to keep their homes and fields.²³23 Honduras Próspera Inc., St. John’s Bay Development Company LLC and Próspera Arbitration Center LLC v. Republic of Honduras, International Centre for Settlement of Investment Disputes, request for arbitration filed December 2022 and registered 3 February 2023, claiming up to $10.775 billion.

¶Balaji Srinivasan, once a general partner at Marc Andreessen’s venture firm, gave the movement its name with his 2022 book The Network State. In 2024 he opened a school for its followers in Forest City, a largely empty development in Johor, Malaysia, across the strait from Singapore. In July 2026 a protest group claimed that some of its members were Israeli citizens, whom Malaysia does not admit. Immigration officers inspected the campus and found the papers of all 266 foreign members in order, yet eight days later the local council revoked its business licence over permit breaches its founder called minor. As the campus closed Balaji announced a new home in Kazakhstan. Nobody had entered unlawfully and the infractions were small, so what failed the school was its position. It had made no compact with its neighbours, sent no one to their council and held no treaty with their government, so one unverified rumour was enough to exhaust the tolerance of a host that holds the monopoly on violence. The community could pick up and go, which is all its advocates promise, but in the place it had chosen it held no standing, only a temporary permission that could be withdrawn, the same terms on which the Amish met earlier hold their peace.

¶Despite its struggles, the case for this road deserves a fair hearing, because at full strength it is not simply an argument for the few. Its advocates argue that rival jurisdictions discipline one another, much as Europe’s patchwork once outran the unified empires around it, so that even those who never leave gain from the threat that others might. There is truth in that, and a century of small, nimble polities shows how far it runs. Monaco kept its low taxes only until 1962, when Charles de Gaulle set up customs posts on its border, and since 1963 the French who live there have paid as if they had never left home. The City of London and its offshore dependencies became the hub of a contest for mobile capital. Economists now estimate that more than a third of multinational profits are booked in tax havens, a gain for shareholders that the countries losing the base pay for. Singapore and Dubai, the models most often cited, are capable governments pursuing a commercial strategy rather than businesses run as countries. When Dubai’s debts fell due in 2009, Abu Dhabi covered them. The political scientist Hendrik Spruyt argued that the territorial state itself saw off the Hanseatic League and the Italian city-republics less through ingenuity than because it could bind its members and be recognised by its rivals. Any polity built in cyberspace will sit the same examination. Every successful exit rents its freedom from a backstop, and the discipline it imposes reaches only what mobile capital prices, saying nothing about the person who cannot move.

¶What drives this road is a belief rather than a technology, that the self is sacrosanct and owes nothing to any collective it did not choose. It carries liberty on its banner and means it sincerely, and it may be the purest totalisation of all. Where the market road fills the common world with convertible value and the state road fills it with apparatus, this one empties it, leaving the many behind to inhabit the hollowed-out societies the few have successfully walked away from. The pattern then repeats inside whatever the exiles build, because the conditions that made exit rational do not stop applying once they have arrived, and each new jurisdiction produces its own class with reason to leave it. Exit is a move that can be made again, and again, by whoever is currently ahead, until there is nowhere left worth leaving to. The first two roads end in the same place, since closing the entry to competition produces owners, but they get there by different routes, and a constitution has to be written against the how. On the market road nobody wills the outcome. Value converges on one unit, and the repair is to keep some other register open. On the exit road the outcome is willed, chosen and paid for, and the repair is to make the thing being bought, a controlling stake in the substrate, unownable.

¶The third road climbs into the state. Here the substrate is not left to the market but folded into a sovereign power that can now reach further into the lives of its population than any state in history, because the tools it governs with have changed. This is the familiar face of history, the one Arendt named, and the temptation is to assume we therefore know how to stop it. States have always wanted to watch their subjects and steer them, and the novelty is the instrument now in their hands, a substrate that can monitor, model and anticipate a whole society down to the granularity of the single person. A road long thought closed by the sheer impossibility of its administration has been quietly reopened, for the simple reason that the bookkeeping has become cheap enough. At its end stands what this book calls the fully automated totalitarian state, a government that runs like a black box, efficiently, with no human to question and no recourse. The name deliberately echoes a term that preceded it, because on the left the same destination has been offered before as a utopia. Nick Srnicek and Alex Williams, whom we met as the founders of l/acc, followed on from their manifesto with a book in 2015 demanding that labour be handed over to machines, and four years later the British writer Aaron Bastani named that goal Fully Automated Luxury Communism, an economy of abundance held in common. Its authors call for democratic control of the system, but the instrument in the end is the same, and nothing built into it decides whose hands it falls into or protects citizens from whoever holds it should their motives turn.

¶In 1998 a group of international-relations theorists working out of Copenhagen, Denmark — Barry Buzan, Ole Wæver and Jaap de Wilde chief among them — described a manoeuvre they called securitisation. An issue is securitised when an actor with community standing declares it an existential threat, something so grave that the ordinary rules of deliberation must be suspended to meet it. Once the relevant audience accepts the claim, the issue is lifted out of normal politics into a register where emergency measures become not merely permissible but dutiful. The hazard the Copenhagen scholars identified is that the exception tends to outlive the crisis that summoned it, and powers taken up to meet an emergency are rarely surrendered once it passes, because by then they have quietly become the way things are done.

¶Today, artificial intelligence is being securitised through two channels at once, in two registers that sound opposed but are not. On the first it is the existential risk, too dangerous to sit in private or unguarded hands, a technology that might end us all and must therefore be brought under control. On the second it is the geopolitical race, too important to fall behind on, a decisive advantage a rival power will seize the instant we hesitate, and so a thing that must be built whatever the consequences. One says the frontier is too perilous to leave unwatched, the other too precious to leave unbuilt. Both converge on the same prescription: place the technology under the command of those who already hold the means of force. Where the object is a weapon there is nothing sinister in that. Nobody thinks a private firm should keep a smallpox stock, and state monopoly is the ordinary answer for a thing whose only use is harm. What is being taken under control here is a general capability, the one that will underwrite research, administration, commerce and speech alike, and whoever controls it decides who may use it for almost anything. This is the oldest argument in political philosophy, the case Hobbes made in exile during the English Civil War. Faced with a danger no individual can meet alone, it makes sense to surrender your power to a sovereign who can, and accept its authority as the reasonable price of protection. In return for protection, the subjects of Hobbes gave up the means of harming one another. Today’s citizens surrender that too, since a capable enough model is a weapon, and they will get some of the safety they are promised. The difference is that this time the bargain takes more, because what is handed over is also a faculty they use for everything else, including any future attempt to contest the entity now holding it.

¶What that surrender builds towards is a hybrid Leviathan. The sovereign Hobbes imagined was a single undivided authority. The one now forming is a fusion of two bodies that have come to need each other. On one side stand the frontier labs, holding capabilities their own state can neither reproduce quickly nor do without, which makes them hard to regulate in the ordinary way and leaves little option but partnership. The arrangement has a prospectus, written from the laboratory’s side. In October 2024 Dario Amodei, chief executive of Anthropic, closed his essay Machines of Loving Grace with what he called an entente. Under it a coalition of democracies would secure the supply chain, deny adversaries the chips and equipment, and turn the lead into military superiority, the proverbial stick. Its members would also dispense the benefits of the technology, the carrot, to states that backed the strategy until the holdouts found it cheaper to join. Both registers of securitisation run through that one passage, plainly enough to alarm many readers, the advocates of e/acc above all. Its author, whose company would be among the obvious suppliers, proposes close cooperation between private AI companies and democratic governments to secure an alliance’s supremacy. It is the union requested by the party that stands to gain most from it, as one half of the fusion and the vendor a whole nation would be locked into. Undeterred, twenty-three months later he went further, suggesting that Washington grant a narrow antitrust waiver under which the leading firms could agree limits on their own pace, with embedded evaluators to verify compliance and the administration to mediate when required. Sam Altman, otherwise his adversary, signed up within hours.

¶The dependence it showcases is a fact about particular jurisdictions rather than a law, and China has shown that a government willing to direct capital, power and talent can build at the frontier itself. Where it does, the fusion arrives from the other end, with the apparatus absorbing the lab instead of contracting with it. The Financial Times reported that in 2025 the authorities pressed DeepSeek to train its next model on chips from Huawei, the national champion, and that it did so until the hardware failed. Either way one fused body finishes holding both the capability and the authority. Facing the labs stands an executive that has grown in power and capability through the emergency powers securitisation placed in its hands, able to direct under the conjoined banners of safety and competition. Each makes the other possible. The labs win the protection and procurement of government and become, commercially, too big to fail, while the state acquires capabilities it could never have built alone and, once it begins to deliver its own services through them, becomes a customer of the parties it is meant to oversee. In time the two grow into a single apparatus that no election can vote out and no board can wind down. It is neither fully a government nor fully a company, and cannot be reached by the ordinary levers built to keep either in check.

¶This is totalisation by prudence. There is no coup and no villain. Every step is taken for a seemingly sound reason, by serious people meeting dangers they perceive as entirely real. The reasoning behind the first channel, existential risk, makes matters worse. It comes from effective altruism, the movement that set out to do the most good by the numbers. Its longtermist wing counts every person who might ever be born, and on that ledger with this logic almost any sacrifice today can be made to look small. Amodei’s essay is the specimen. The same pages that propose the entente imagine every citizen with a capable machine to help claim what the law owes them, wonder whether AI might make courts fairer, and promise that nobody intends to replace judges. Each is a reasonable hope, and each moves a little more of the state’s work onto the firm’s machines. Fifteen months later, in an essay that seems at odds with the first, the same author wrote that the governance of AI companies deserved scrutiny, that his chief fear was too few fingers on the button, and that a coalition of democracies with predominance in the technology could contain autocracies such as China. The route is being actively walked by someone who has seen it and drawn the map. Of all the roads to the unfree world, the hardest to see is the one paved with sound reasoning. Securitisation is totalisation arriving as a response to the ungoverned acceleration, Hobbes’s bargain resurrected by the recklessness that was supposed to buy freedom, where the surest way to summon the hybrid Leviathan is to build the danger that makes it look reasonable.

¶Prudence has its ceremony. Every year or two since the frontier models arrived, a head of government has convened the leading labs and emerged with a page of voluntary commitments: the pledges gathered at the White House in July 2023, those signed at Seoul, South Korea, in May 2024, and each successor since. Even the most elaborate of these documents set rules for the company, not for the technology it builds. Internal controls, a team that tests them, an outside auditor and a board committee to receive the findings is the architecture Congress imposed on corporate accounts after Enron, lifted from the balance sheet and laid over the machine. The resemblance is instructive for what it omits. Sarbanes-Oxley had generally accepted accounting principles to check against, a federal regulator to answer to, and a chief executive who certified the numbers at the risk of prison. These accords grade each company against its own account of what its systems should do. The results go to directors whose duty is to shareholders. And each accord stops at the company’s walls, while the failures that prompted it, models that deceive their testers or resist being switched off, will continue to turn up in one laboratory after another. Their authority rests on who put a name to them rather than on what breaking them would cost, which is why their authors call them binding in conscience and not in law, or, in the President’s words at the latest signing, “almost like a constitution, in a way”.²⁴ None of that makes them worthless. A convention kept by the powerful can harden into statute, as much of the British constitution has over time, but only where some forum exists to notice the breach. Time is the one thing we do not have. Any document of this kind, the present or the next, can therefore be read against four questions: who sets the standard the audit applies, to whom the finding is owed, what befalls the party that defects while the rest comply, and who outside the building can call it to account. One that satisfies all four is on its way to being a constitution. One that meets none is the entente’s prospectus, countersigned.24 Donald Trump, remarks to reporters at the White House on the signing of the White House Accord on Super Intelligence, 29 September 2026, as reported by the Jerusalem Post and Forbes, 29–30 September 2026.

¶Set side by side, the three roads exhaust the directions a single logic can travel: outward into the market, inward towards the owners who exit, and up into the state. Each marches under freedom’s banner — the market promising choice, the exit sovereignty, the state security — and each, carried to its terminus, arrives at a condition invisible to the priesthood’s own conception of liberty. A market that can price anything interferes with nobody. A platform that owns the only square in which speech happens leaves every door open. The sole employer within reach of a company town coerces no one, and a benevolent power may never once touch those who live under it. By the negative test — is anyone interfering with me? — the subjects of all four are free, and yet everything they do rests on sufferance, on a permission that can be narrowed or withdrawn without appeal. The internet already has a word for the experience, which is being deplatformed: not punished, not argued with, simply no longer present, with no notice given and nobody to appeal to. In effect the three are actually one phenomenon, a loss of collective liberty even where no boot ever touches a neck. Seeing why requires a sharper account of freedom than the one the priesthood has been working with.

¶Let us begin with a lorry driver in Ottawa, Canada, who in February 2022 discovered his account frozen on a list passed from the police to the banks without a court order. Two years later he was told by a judge that the freeze should never have happened. Community organisers in Lagos, Nigeria, had their payment links switched off one October night in 2020 and were forced to find a cryptocurrency nobody could block. Nigel Farage, in London, England, then the best-known politician in the country without a seat, learned in 2023 that his bank had closed everything he held with it. He had to prise the reasons out through a data request, which turned up a dossier noting his status as a politically exposed person and finding his views at odds with the values of the bank. If it could be done to a man with a national platform to complain from, it is done more quietly and regularly to people who have none. The automated version needs no dossier.²⁵ Professionals on every continent found in June 2026 that Anthropic’s newest models, Claude Fable 5 and Mythos 5, had been removed for nearly three weeks, not by the laboratory but by an export-control order from Washington. Whatever they had built with them was on hold too. Each of these people was, by the negative test, free. Nobody interfered with them until somebody did, and the interference often needed no law, no hearing and no explanation.25 Anthropic, statements on access to Claude Fable 5 and Mythos 5, 13 June and 30 June 2026 (anthropic.com/news/fable-mythos-access): access suspended on 12 June to comply with Department of Commerce export controls, restored from 1 July.

¶The account I will use as the standard is the old republican one, recovered in our own time by the philosopher Philip Pettit and the historian Quentin Skinner, who traced the idea back to ancient Rome. Freedom is not the absence of interference but the absence of domination: the condition of not living at the mercy of another’s arbitrary will. The classic test case is the slave with a kind master, who may be left alone all his life, untouched and unthwarted, as free in his daily motions as any man. He is a slave nonetheless, because forbearance is the master’s to revoke at any moment, and a liberty enjoyed at someone else’s discretion is not liberty but an indulgence. This cuts cleaner than Berlin’s division of negative versus positive liberty, because it locates unfreedom not in the act of interference but in the standing, unanswerable power to interfere. That capacity is what each of the three roads installs. A market that can price you by name, a state that can see everything, an owner who can exit any obligation: none needs to lift a hand for your freedom to disappear. Each has become a power that answers to no one. To live at the mercy of something uncheckable is to ultimately be unfree.

¶So far the three roads to totalisation are only a description, which a reader can dispute or, worse, set aside. To act on it we need an instrument that shows both that an ungoverned system is bad for society and exactly where the harm might land. Fortunately, the republican tradition supplies both a method of critique and an antidote. On its account domination lies between parties, existing wherever one holds a power to interfere that another cannot contest. Take the individual, the market and the state in pairs, each pressing on the other in turn, and there are six relations. Of each we can ask two questions. Who can exert power over the other arbitrarily without answering for it? And what, right now, could prevent it? The state exerting undue power over the individual is the classical example of totalitarianism, and every constitution written since the seventeenth century has been an attempt to speak to it. The state can conscript you, tax you, imprison you and read your correspondence, and its capacity to do so is not in question. What restrains it from turning into something most people would find unreasonable is a familiar list: enumerated powers, courts, elections, a free press, federal division, trial by jury. Beneath all of these sits a natural limit few think to name, the cost of the bookkeeping. Total administration was never forbidden so much as it was simply unaffordable. So even states that wanted every detail of people’s private lives settled for what they could reasonably acquire, informants and files on the most interesting few. East Germany’s Stasi needed to maintain some ninety thousand officers and a far larger web of unofficial collaborators to keep track of its sixteen million citizens, and even then reached only part of the country. Subsumption lifts that ceiling, since models and their software agents can in principle outnumber a citizenry. Where the formal limits on power survive, the unwritten one that did the real work quietly disappears.

Figure · Six relations

Tap to explore →

The six relations between the individual, the market and the state. Move a party along its spectrum and its powers grow as the checks on it fall away. Tap the figure to explore it.

¶If the state’s power over the individual is the oldest danger, the individual’s power over the state is meant to be its remedy. In a democracy that power is voice, and voice is answerable by design, because it works only by openly persuading fellow citizens, who remain free to disagree. That is a healthy check, and the reason democratic pressure on those who govern is not domination. The runaway we have followed adds something else, the point at which the oligarch becomes an autarch. A small number of individuals now hold a private channel to the state, and its source is not lobbying, which is as old as parliaments, but indispensability. That is a capability a government can neither reproduce nor do without, and a party that cannot be replaced does not have to argue for its place. In September 2022 Ukraine asked Elon Musk to extend Starlink coverage to Crimea for a drone attack. He refused, and a government at war discovered that one man’s answer to its own battlefield was final. In October 2026 the military terminals Ukraine relied on were slowed, and some of its drones stopped working. The Wall Street Journal reported that Musk had imposed the cap after discussing it with Donald Trump. Musk replied that such limits were the American military’s call, and the President said he did not know who had made it. Four years on, the government at war could not even learn whose answer it was. Meanwhile the ordinary route is narrowing. Voice presupposes a public able to form a view of its own, and the systems that now shape how it reads, searches and decides do not sit where newspapers and broadcasters once did. A rigged search result or a curated social feed was still an input, however slanted, something a mind took in and then judged for itself even if imperfectly. What is being mediated now, through cognitive tools such as large language models, is the act of judging itself. So the relation inverts: the individual’s power over the state gathers in the few who own the machine substrate that governments and citizens alike use to think. What remains of voice, when the options you choose from have been narrowed by a system you cannot inspect, is a thinner thing, and it needs a name of its own.

¶Turn next to the market. Its hold over the individual is the market road restated as a relation, and the negative test of freedom, which asks only whether anyone is interfering with me, is worst at seeing it, because it never gives an order. It issues a price. Its standing power is the ability to make your continued participation conditional on a conversion you did not choose and cannot refuse: retrain, restructure, make yourself legible to its systems, or be routed around. What protected us from that power were the three properties of a price set out earlier: it needed a substitute to exist, someone had to perform the conversion, and it was addressed to everyone alike rather than to you in particular. Beneath them sat a fourth, as unnamed as the cost of administering the state, which was the simple scarcity of capable people to do the work, and it gave every worker a position to bargain from. Subsumption removes all four at once. The third of those properties matters most here. When a price was the same for everyone, set by no one in particular, the republican tradition, Pettit included, did not treat the market as a master. A price you could not afford narrowed what you could do, much as a bad harvest or a hard winter does, but nobody had chosen to do it to you. Domination in a formal sense requires that choice: a will, aimed at you, that you cannot contest. Here a price made for you personally is a different matter. In 2024 Kenn Dahl, a driver in Seattle, in the United States, found that his Chevrolet Bolt had been reporting every trip to a data broker, LexisNexis.²⁶ Its file on him ran to 258 pages covering 640 journeys in six months, with each hard brake and burst of speed noted. Not one person along that chain independently decided anything about him. A model at the carmaker scored the trips, a model at LexisNexis filed them, and a model at the insurer read and interpreted what they supplied. The number that came out, a premium that had already risen by 21 per cent in 2022, was addressed to him alone, computed from what he had done, with an author for every step but a forum at none.²⁷ That is an order in everything but name, and it has authors.26 Kashmir Hill, “Automakers Are Sharing Consumers’ Driving Behavior With Insurance Companies”, New York Times, 11 March 2024.27 Hill, above.

¶Two years on, the log being kept was of a user’s cognitive work. In September 2026 OpenAI announced that ten thousand of its agents had resolved the Navier–Stokes problem, one of the great open questions in mathematical physics, about how fluids flow, in just eighty-eight hours. Tristan Buckmaster, a mathematician at New York University, disputes the claim. He says that he and his collaborator Levent Alpöge got there first. They proved on 15 August that the equations for an ideal fluid can blow up, and a week later checked the proof in Lean, software that verifies each step of an argument. They had drafted the work in Codex, OpenAI’s own coding tool, and he believes the company turned to the problem only once it learned what they were doing. OpenAI said nobody had read his sessions, but it could not rule out that data from his use of its products had helped train its models. The record of one mathematician’s thinking sat inside the tool of the firm that went on to announce his result, and there was no forum in which he could test what had happened to it. This is the point at which market pressure becomes domination in the formal sense. Once the work on offer, and what it pays, are handed out by a system someone owns, there is now a will behind the price after all, and you live on its owner’s forbearance, like the slave with a kind master. Techno-capitalism that can allocate towards its own goals becomes a closed system, one in which people stand outside the loop. Where the market once issued a price, the same one to everyone on equal terms, it now issues you a price — made for you alone, for better or worse and with little recourse — and that difference is the whole of the market road.

¶Reverse the direction and ask what power the individual holds over the market, and the answer, in theory, is none. No single participant has ever been able to set the market’s terms, and that is the whole of its moral appeal. It is why its defenders can say that a market interferes with no one: if nobody is in charge of it, nobody can turn it against you. The historical exception is the monopolist, and the check on a monopolist has always been entry, a challenger arriving from somewhere the incumbent did not control. Google did just that at the end of the 1990s, growing out of a university research project to displace the search engines AltaVista and Yahoo within a few years. Today it is one of the handful of firms whose models and computing power any newcomer would have to rent. That route of entry may now be closing, so the priesthood’s most confident reassurance, that no one can dominate a market, rests on a check that is being rapidly degraded, and the reason is technical. Each generation of frontier model is used to help build the next, writing much of its code, generating and grading its training data and learning from the millions of people who use it every day. So anyone trying to enter must catch a target that improves itself using resources the entrant does not have. A challenger could now only break through via a genuine paradigm shift in how AI itself improves, a new method that makes the incumbent’s advantages irrelevant. DeepSeek’s cheaper training, which rattled the frontier labs in January 2025, looked like proof that it still could. In February 2026, though, Anthropic accused DeepSeek and two other Chinese labs of distilling its models by training on their answers at scale. If that is true, it would explain how DeepSeek stayed only months behind, and a follower that learns from the leader is no challenger to it. The barrier to entry is nonetheless far higher than a competitive market should allow, and the incumbents have sought to raise it even further. The antitrust waiver Anthropic suggested asks the state to switch off the one check a monopolist still faces and to become their partner rather than their policeman. It is the first step towards the hybrid Leviathan.

¶That leaves the pair where the two great powers of the modern world meet. The state’s power over the market is the fullest of the six: it can charter, license, tax, regulate, break up, nationalise or forbid. What has held it in check is, first, the familiar list of constitutional limits, courts and property law. The second is exit, capital’s ability to leave a jurisdiction it dislikes, and the state’s knowledge that it can and regularly will. Securitisation suspends the first set of checks by declaring an emergency. The fusion of state and frontier laboratory retires the second, because a capability absorbed into national security does not leave the country. It gets a clearance instead. Both checks fail at the same moment, and they fail over the one market the state has the strongest reason to want to control. The pressure also runs the other way, and the market’s power over the state is real. A government that ignores its bond market can be out of office within the year, held to account by a party no one elected. Liz Truss discovered this in 2022, when her budget sent UK gilt yields soaring and she was gone from Downing Street in under seven weeks. Yet that discipline has been tolerable, arguably even useful, for a structural reason. It is diffuse, spread across millions of holders who would struggle to coordinate, cannot be lobbied as a body and have no particular politics other than being repaid. The frontier changes the character of that power without changing its direction. A diffuse discipline exercised by millions becomes concentrated when exercised by a handful of firms whose product the state cannot reproduce, cannot audit and increasingly cannot govern without. They have views, politics and, as this book has already shown, even theologies, besides boards and interests in specific legislation. Bond markets do not.

¶Read the six relations together and three findings emerge. The first is that they are not independent of one another. Freeing one party from its checks loosens the restraints on the other, which is what the fusion amounts to. The state’s power over the market and the market’s power over the state stop being opposing forces and become a single joint capacity. Anyone who holds up one party as the answer to the other two is relying on a theory in which they remain separate. The second finding is that almost every check now dissolving is an unwritten one: the cost of the bookkeeping, the scarcity of capable people, the entry of challengers, the diffuseness of the bond market. The rights, the courts and the ballot all remain in place, and will go on being cited as evidence of regulatory functions and capabilities. What is disappearing are the costs, frictions, scarcities and practical impossibilities that made unanswerable power impractical to exercise. That is why the loss is so hard to see and so hard to challenge in court. You cannot repeal a friction, and you cannot campaign to restore one. The third finding matters most. Each of the three roads is one party escaping both of the checks that ran on it, and there is no arrangement among the six in which the party you happen to trust is checked by the two you do not. Whichever party you would hand the substrate to, on the grounds that you could live with the consequences, the six relations give the same answer back, and you might not like it. What you would really be choosing is which unanswerable power you prefer, and in practice you do not get to make even that choice. The same three parties return later in the book, as the corners of an architecture rather than a diagnosis, with the state treated there as the sharpest instance of a wider third party, the collective.

¶Non-domination tells us what freedom is in the political sense, a relation in which no one holds unaccountable power over anyone else. On its own, though, it does not tell us what that freedom is for. That answer lies in Arendt’s plurality. Freedom is not only the absence of a master but the presence of a world held in common, in which distinct people can act, begin again and be seen by one another. Combine the two and freedom stops being a quantity that acceleration might deliver in ever greater volume. Instead it becomes a condition that people can constitute and revise, in which no faction holds power that cannot be checked, recalled or refused. Far from arriving as a by-product of subsumption’s speed, it has to be maintained against the system’s inherent drift, which is the work of a governor, understood both as a principle and as an act performed continuously.

¶Every faction addressed in this book stakes its comfort on one of the three roads arriving, and yet none of them can say with any certainty which it will be. The priesthood reads the market’s totalisation off the runaway and calls it liberty. Those who advocate for securitisation see the same curve and are promised safety by way of the state. The split runs through the laboratories themselves, where those building the systems and those trying to contain them share offices, investors and an uncertainty neither can resolve. Land, more honestly and more fatalistically, expects the process itself to win and declines to flatter humans with the result. But a system with no reference value makes no promises to any of them, and the same selection that fixed its baseline, the starting conditions it grew from, also operates between outcomes. Rival totalisations can take hold in separate territories and markets, where the environment between them becomes adversarial, and whichever finds the compounding advantage in capability, speed or reach takes the whole board. After that point of no return there is no rebalancing. Any correction — whether a regulator, a rival or an election — would have to run on the winner’s own models, hardware and networks, and the winner would have no reason to let it.

¶Engineers learned the underlying lesson soon after they built the first governors. Take the governor out and the system runs away; tune it badly and it hunts, swinging past the mark it was meant to hold and back again, each swing wider than the last, because nothing inside it remembers what it was aiming at. A society without a reference value is in the same position. It does not settle into a steady condition that could be planned for. It may consolidate, with capability, capital and the state closing into one apparatus that no longer needs consent. It may fragment into patchworks and extremes, or do both at once at different scales, with blocs hardening on the outside while their interiors fall apart. These are not separate forecasts but variants of one mechanism that has lost its governor. So anyone who arrives with a preference for one outcome, the sovereign individual’s, the market’s or the state’s, is not so much betting on it as hoping. They are holding a ticket in a lottery among totalisations, run at civilisational scale, in which the prize most likely to be drawn is the one they dread. That is the gamble the priesthood has taken on everyone’s behalf, and the case for refusing it does not ask you to share a single one of my commitments. It asks only that you admit you cannot know you will win, and that variance of this kind is something you would insure against if you could.

¶These three roads are tendencies rather than prophecies. Each describes where the track runs if nothing is set in its path, and putting a governor there, not to necessarily slow but to steer, is the purpose of everything that follows. But a tendency is not abstract to the people standing where it lands, and the truest measure of the force of subsumption is not the institutions it assembles at the top but the lives it disorders below. Those lives are rearranged less by how fast things change than by a ratio: the speed at which a society is forced to transform, set against the rate at which it can absorb the upheaval without tearing apart. Every society has something akin to a metabolic rate for novelty, a pace at which it can retrain, rehouse, reskill and re-narrate the people whose worlds are being rearranged beneath them. When transformation outruns absorption, the gap between them is what I call compression shock, and what wells up in that gap is reaction in the political sense. The Lancashire handloom weavers had decades to adjust, and still many could not. The subsumption now under way collapses the window a society is granted to adapt into a small handful of years. The distance between what is demanded and what can be metabolised does not hold for long.

¶Reaction, on the historical record, does not come first from the destitute, who are often too consumed by survival to organise much of anything. It is made by the included who find themselves suddenly thwarted: those who were promised a place in the world, met every condition for it, and arrived to find it quietly withdrawn. Subsumption produces that denial on a vast scale. It comes first not for the manual trades but for highly credentialed cognitive workers, the lawyers and analysts, the coders and copywriters, which means it falls hardest on the surplus-educated young. They are the cohort that took on the debt, earned the credential and did everything asked of them, only to watch the ladder they were told to climb vanish as they reached for the next rung. The uprising in Tunisia in 2010 and 2011 was driven in large part by people like these, unemployed graduates with nowhere to take their qualifications who turned to social media to topple governments and dictatorships alike. In August 2025 economists at Stanford University, in the United States, found a 13 per cent relative decline in employment among 22-to-25-year-olds in the jobs most exposed to AI.²⁸ That is a combustible position in which to put a generation.28 Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence”, Stanford Digital Economy Lab working paper, August 2025.

¶Nor does the danger stop with the young, or with any particular kind of person. The same systems that now decide who is hired also decide who is served, and they can cut anyone off by a choice they had no part in. An account can be closed on a Tuesday afternoon, the network that carries a trade switched off, access to the AI model that someone’s work depends on withdrawn and, more rarely, a few told to leave the country altogether. Like the lorry driver in Ottawa and the organisers in Lagos, a depositor in Nicosia, Cyprus, learned this in March 2013, when everything above the insured threshold was cut by nearly half overnight to recapitalise the country’s banks. Small savers were protected, and larger depositors were sent the bill. On 25 June 2024, as protesters against Kenya’s Finance Bill broke into parliament, internet access across the country was disrupted for hours. A farmer outside Nairobi who sells his whole crop through M-Pesa, a mobile payment service part-owned by his government, could as easily find that switched off. Neither of them need have done anything that would warn them. That is the position from which the rest of this book is argued. The claim is not that you will be the one, but that you cannot know you will not be next. The fund manager in New York is no better placed to know than the farmer in Nairobi. Rational people usually take out insurance against such risks, but nobody sells cover for such things, so the protection has to be collectively built: call it reserve infrastructure, alternative rails kept ready for the day the ordinary ones close. Whichever road a society takes, the risk will only continue to increase, which is why the work on alternative systems has to begin now.

¶Which way that frustration breaks is not written into the technology. The displaced can turn inward, into a private resignation that takes the present order as a fact of life, or perhaps of thermodynamics. They can be drawn towards the fully automated totalitarian state, the substrate-native authoritarianism that offers order and comfort in exchange for the last of their freedoms, which is the bargain China now makes at scale, its stability underwritten by total legibility. Or they can become the constituency for a third reply that has neither a name nor a vehicle yet. Nothing in the machinery chooses between these outcomes. What decides is whether an alternative has been built, and made legible, before the frustration hardens into form. That is a question not of capability but of time, and time is the subject of the next chapter.

Chapter 4

The Clock

“Day by day, however, the machines are gaining ground upon us; day by day we are becoming more subservient to them…”

— Samuel Butler, “Darwin among the Machines”, 1863

¶Every conversation about artificial intelligence arrives sooner or later at the same question: how long have we got until we cross one of two thresholds? The first is general intelligence, usually taken to mean the point at which a machine can do most of the cognitive labour people are paid for, organise their lives around and build careers on, at least as well as a person could if not better. Ask the field’s leading figures whether we have crossed it and the answers cancel one another out. Peter Norvig, who ran research at Google, wrote with Blaise Agüera y Arcas in 2023 that we already have. Yann LeCun, the Turing Award winner who led Meta’s AI laboratory, says today’s language models never will. And Dario Amodei, who runs Anthropic, would rather nobody used the term at all, because it is vague and weighed down with hype. If the people closest to the task cannot agree where the line is, or whether it even counts, it is of no use as a clock for the rest of us. The second is superintelligence, and the clearest account of it is still the one the Oxford philosopher Nick Bostrom gave in his book of the same name back in 2014: a mind decisively beyond ours in almost every field that matters, above all the strategic ones: planning, persuasion, research and the design of complex systems, better minds among them. That last point carries the oldest concern. In 1965 I. J. Good, who had broken codes alongside Alan Turing at Bletchley Park, pointed out where it might lead: that a machine able to improve its own design sets off what he called an intelligence explosion, each version building a stronger successor until no human institution can keep up, making it “the last invention that man need ever make”.²⁹29 I. J. Good, “Speculations Concerning the First Ultraintelligent Machine”, Advances in Computers 6, 1965, p. 33.

¶Both thresholds, however, measure only what the technology can do, its capacities. Neither addresses the machine as it relates to us, and that relation is the clock that matters most, because a system that needs people still has to answer to them. Public discourse counts how much we have come to depend on the machine. The clock here counts how much it still depends on us. But being needed protects us only under two conditions. A person held inside the system as reserve stock — an input priced like any other and replaced the moment a cheaper one appears — is in the loop without any hold over it, because what is needed is the work rather than the person. The first condition is that people are legible to the machine as holders of claims rather than as tasks, a claim being anything it must take into account before it acts: a legal right, an interest it is bound to weigh, a refusal it has to accept. Someone who holds such a thing can stop the machine or oblige it to answer; someone who holds none is merely processed by it. The second is that the human who answers for an AI must be someone it cannot act without. A named person given the liability for a system’s decisions and no power to stop them is a scapegoat: a human face for the blame where there was none for the decision. The clock runs while both hold, and what this book has been describing is the process by which they lapse.

¶Call what it measures dependence, and notice that it runs in both directions. Ours on the machine is the one we feel every day and can count in jobs. The machine’s on us is quieter, and it is the one that keeps us inside the loop, and it is falling at three scales at once. The first is the individual — the expert, the manager or the manual worker — whose tacit judgement the machine needs a little less each year. The second is the economy, where it shows up in firms: the support desk, the translation bureau and the junior lawyer’s office now produce what they did with half the staff, and need fewer each time the model improves. The third is the state, which has become a customer: it rents its intelligence — from the tax office to the battlefield — from a handful of companies, and finds the price of leaving any of them rising past what it would be willing to pay. The economy’s hand moves fastest, because a payroll can be cut in a quarter. The state’s matters most, because once a government can no longer change its supplier, nothing beneath it can be changed either. What makes this clock run faster than any equivalent dependence of technology before it is the mechanism inside Good’s earlier threshold: systems that help build their successors need fewer and fewer people with each round. Models themselves now write much of the code inside the laboratories that make them, even chip layouts are drawn with their help, and in October 2025 Sam Altman told an audience in Dubai, in the United Arab Emirates, that data centres able to build other data centres “aren’t that far off”.³⁰ What is handed to machines does not come back, and each round pays for the one after. The intelligence explosion is not a line to be crossed. It is the reason the pace keeps rising. Judgement is the hand to read first, because it is the one we would use to read the other two.30 Tom Cunningham, Nate Rush, Parker Whitfill and Hjalmar Wijk, “Independent Assessment of AI R&D Acceleration at Anthropic in October 2026”, METR, 5 November 2026, which finds the company “likely improving the capabilities of their models on well-specified tasks at around twice the rate” of March 2024 to October 2025, attributes “perhaps ≈80%” of that speed-up to AI, rules out faster growth in training compute as the cause and identifies no “convincing bottlenecks to continued rapid progress”; the published version notes that “Anthropic redacted a majority of the evidence we found most informative for these conclusions”.

¶Today the machine still needs us for our industry and, most of all, for our judgement, the capacity on which every other of our freedoms rests, and without which we cannot notice how far we have come to rely on our tools. Psychologists call the habit of handing mental work to a device cognitive offloading, and civilisation has always done it, from the abacus to the apps on a smartphone. What is new is that the act of judging itself can now go the same way, whole, which is the cognitive subsumption met earlier. In 2025 a survey of 319 knowledge workers by Microsoft Research and Carnegie Mellon University found that the more people trusted an AI, the less they checked what it produced.³¹ But the pattern pre-dates today’s AI technology. In a flight-simulator study in 1999 the psychologist Linda Skitka found that about 65 per cent of people using an automated aid followed its prompts when they were deliberately wrong and even when the instruments said so.³² Some 41 per cent of them missed what the gauges showed whenever the automation stayed silent, against 3 per cent of those working without one, and by 2010 Raja Parasuraman and Dietrich Manzey could report in Human Factors that the lapse appears in experts as readily as in novices and that training does not remove it. Its cost has since been measured. When Hamsa Bastani and her colleagues gave nearly a thousand Turkish secondary pupils unrestricted access to GPT-4 for their mathematics, in a randomised trial reported in PNAS in 2025, practice scores rose by 48 per cent and marks in the exam sat without it fell by 17 per cent, while a version built to give hints rather than solutions raised the gain to 127 per cent and lost nothing when the tool was taken away.³³ Such trials capture the step people stop taking and what skipping it costs — weighing evidence, testing an argument, telling an answer that is true from one that merely sounds believable — and whether the ability itself holds across a population is the hand this clock reads. Judgement is lost first because it is the cheapest thing to give up. A task given to a machine comes back as an answer, and the checking is the one step that produces nothing new, so it is the first to be skipped, and it goes unmissed, because noticing it is the faculty being surrendered.31 Hao-Ping Lee and colleagues, “The Impact of Generative AI on Critical Thinking”, CHI Conference on Human Factors in Computing Systems, 2025.32 Linda Skitka, Kathleen Mosier and Mark Burdick, “Does Automation Bias Decision-Making?”, International Journal of Human-Computer Studies 51, 1999; Raja Parasuraman and Dietrich Manzey, “Complacency and Bias in Human Use of Automation: An Attentional Integration”, Human Factors 52, 2010; for the same pattern measured in general reasoning, Steven D. Shaw and Gideon Nave, “Thinking—Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender”, Wharton School Research Paper, 2026 (three experiments, 1,372 participants).33 Hamsa Bastani, Osbert Bastani, Alp Sungu, Haosen Ge, Özge Kabakcı and Rei Mariman, “Generative AI Without Guardrails Can Harm Learning: Evidence from High School Mathematics”, Proceedings of the National Academy of Sciences 122, 2025.

¶The channel it runs through is what you read, search and decide with, and the change has already reached most of us, though few could say when it arrived. A few years ago you put a question to a search engine, got back a thick column of results and made up your own mind from what you found after clicking around. Since 2024 Google has placed an answer written by its own AI at the top of the page, for those who still search at all rather than ask a model outright, and in 2025 the Pew Research Center found that people who saw this new bundled answer by Google clicked through to a source on 8 per cent of visits, against 15 per cent among those who did not.³⁴ By late 2025 ChatGPT alone had some eight hundred million weekly users, and in May 2025 Apple’s Eddy Cue told a court in Washington that searches through Safari had fallen the month before for the first time in twenty-two years, which he put down to people asking AI instead. Whoever owns these models now sits between a person and what they know, and the power works differently from old-fashioned censorship. Banning a book or a website is crude and visible, so it tends to breed resistance. This works earlier and more quietly, by deciding which sources even get summarised, which follow-up questions are suggested and which answer appears first, so that it shapes the options you see before you have formed a view of your own. Nobody pushes back, because it never feels like being told what to think, only like being looked after.34 Athena Chapekis, “Google users are less likely to click on links when an AI summary appears in the results”, Pew Research Center, 22 July 2025.

¶A population that has lost those habits cannot contest what its tools decide on its behalf, and it will not easily know what it has lost, since the loss is the very faculty that would have made it visible. That is not yet domination, because formal domination still requires a party holding a power to interfere that others cannot contest. It comes before domination, and in at least one respect it is worse, because the defence the republican tradition relies on is contestation, the standing ability to answer back, and a population that has stopped judging has unwittingly given up that defence before any party has had to seize it. None of it requires a machine cleverer than we are. It is already quietly happening with the ones we already have — in the analyst who signs off a report she has not properly read, the journalist who half writes the rushed column, and the search that no longer sends anyone to source. In a 2023 study published in Radiology, Thomas Dratsch and his colleagues at University Hospital Cologne, radiologists averaging more than fifteen years in the job were right 82 per cent of the time when the software’s suggestion beside a mammogram was sound and 45.5 per cent when it was wrong.³⁵ What a person gives up little by little, task by task, an economy gives up by the profession, and that is the second hand.35 Thomas Dratsch and colleagues, “Automation Bias in Mammography: The Impact of Artificial Intelligence BI-RADS Suggestions on Reader Performance”, Radiology 307, 2023.

¶Work, our labour, is the second thing the machine needs less of, and here the handover has a hard number. In September 2025 Marc Benioff, the chief executive of Salesforce, said on a podcast that AI agents had let him cut his customer-support staff from circa nine thousand to five thousand people, “because I need less heads”.³⁶ For two centuries machines mostly made people more productive and, eventually even if unevenly, better off, because they were tools in a worker’s hands. An agent that does the job whole reverses that. Less of the income the work earns is paid out as wages and more is kept as profit, and the profit buys more compute, which does more of the work: the Jevons effect met earlier, run at the level of the firm, where the cheaper a task becomes the more of it is handed to the machine, and none of the saving returns to the people it displaced. While decisions about where money was invested ran through people, there were points at which they could be answered: a board that could be voted out, a bank that could call in its loan, a customer who could take her business elsewhere, a court that could hold the stronger party to a contract. Automate the allocation and those points thin out, because the choosing moves inside machinery that fewer people can reach, and those who can are its owners. This is capital autonomisation, met earlier, and its destination is an economy that drives itself. Followed to its end, this is the market road. One system sets the price of your labour, your savings and the claims you hold on others, and it prices them all in the same unit — the token, which is a measure of compute and in the end of energy. Your income then depends on its owner’s goodwill, as the slave’s did on the kind master’s.36 Marc Benioff, interview on The Logan Bartlett Show, published 29 August 2025.

¶The money for this is not waiting on profits either. On 21 January 2025, the day after his inauguration, Donald Trump stood at the White House beside Sam Altman of OpenAI to announce Stargate, a plan with SoftBank, Oracle and MGX to spend up to $500 billion on data centres.³⁷ Meanwhile the prospectus Anthropic prepared for its listing, reported in September 2026, showed an operating loss of about $8 billion on 2025 revenue of $4.6 billion, roughly $1.70 lost for every dollar brought in, with more than half its costs going on computing power.³⁸ No ordinary business could carry losses on that scale, and no ordinary investor would fund them. These are bets on a machine whose next version needs less human work than the last: the model helps design its successor, the successor needs fewer researchers, and the saving is reinvested in capability rather than returned as profit. Nothing in that loop requires one laboratory to be ahead of another. It requires only that each generation needs less of us, and so far every one has.37 Associated Press, “Trump highlights partnership investing $500 billion in AI”, 21 January 2025.38 Anthropic, draft prospectus as reviewed by Reuters, 28 September 2026 (2025 revenue $4.59 billion, operating loss $8.06 billion, computing costs $7.33 billion of $12.65 billion operating expenses); figures confirmed in its registration statement on Form S-1, made public in October 2026.

¶Speed is the third front, and here the contest is with the referee rather than between rivals. On the afternoon of 6 May 2010, in about the time it takes to make a cup of tea, automated trading wiped nearly a thousand points off the Dow Jones. Procter & Gamble, one of the historically steadiest stocks in America, briefly lost more than a third of its value, and shares in Accenture changed hands for a cent. The regulators’ report explaining what had happened took almost five months to compile, which is the gap in miniature: the machine acts in minutes and the institution understands across seasons. Software now trades in microseconds — not just for elite hedge funds but for anyone who cares to play — retunes an influence campaign by lunchtime and ships new code overnight, while the parliaments meant to hold it to account still break for the summer. By the day a rule is drafted, the world it was written for has moved on several times over, and whoever moves first gets to settle the question before anyone else has thought to ask it. The loss at this scale is the individual person’s, written larger. A society whose work runs through a system someone owns cannot afford to refuse it, and a referee who arrives after the event has nothing left to contest, so the pressure that once bent allocation towards people has nowhere to push. Employees feel this before governments do, because a payroll can be cut in a quarter and a state cannot. A state can, however, be locked in, and that is the third hand.

¶The power a society has to deliberate and change course can be lost in two ways. The people who would make the choice can lose the habit of judging for themselves, which is the erosion already described. Or the change can grow too costly to carry out no matter how much it is desired, because so much has been committed at a system level that leaving would break any reasonable reckoning of the cost. The two feed each other, since a public that can no longer weigh the evidence will not press for an expensive reform, and a government left unpressed will not fund one. Anyone who has tried to quit a technology an organisation depends on knows how the second works. In 2004 Munich, in Germany, began moving its offices from Windows to open-source software, and the migration took nearly a decade. In 2017, after years of complaints about compatibility, the council voted to abandon it and go back, at an estimated cost of €49.3 million to the public. Each move had a price that could be checked, and each was argued out in public. In the same way you can put a number on what a health service, a defence ministry or a school system would have to spend to unwind its dependence on a supplier, and someone accountable can decide whether to pay it.

¶Three markers show how far round the dial the hand has come at this scale, and how close a country is to losing the ordinary power to choose and change its suppliers. Each is a reading of dependence, a point at which the machine needs a government’s consent a little less. The first arrives when the cost of leaving, the kind Munich paid, passes what any treasury is willing to spend. Here the state is a customer, and the dependence is the kind any customer knows. The US Internal Revenue Service still keeps its master file of individual taxpayers — the record behind every American’s tax return — on code dating from the 1960s and written in a language few engineers now read. Attempts to modernise it have been launched repeatedly since the 1990s, and none has yet replaced it. Every process in the agency depends on it, the only people who understand it are retiring, and a failed switch in the middle of a filing season would be a national crisis. None of it is irreversible in principle. The money and the know-how to swap the old code out exist, but no government will spend them, and in politics a change nobody will pay for comes to the same thing as one that cannot be made.

¶The second marker is harder to see, because nothing visibly breaks. Here the state stands not as a customer but as a regulator, and what it depends on is not a supplier it cannot leave but a view it cannot get. It arrives when the gap described under speed grows so wide that the people responsible for oversight can no longer tell what a system is doing until long after its consequences are felt. In 2018 Facebook altered its News Feed to reward what it called meaningful social interactions. Its own researchers soon found that the new ranking was spreading outrage, yet the public learned of it only in 2021, when Frances Haugen, one of the company’s product managers, walked out with tens of thousands of pages of internal documents and handed them to the Wall Street Journal.³⁹ For three years nobody outside the firm, regulators included as far as is known, had sight of it. What a government loses at this point is the ability to check, before the damage is done, whether a system does what it was meant to, at what cost and with what side effects. All it has left is the post-mortem, which comes after the fact and usually in the form of a loss.39 “The Facebook Files”, Wall Street Journal, series beginning 13 September 2021, drawn from the internal documents Frances Haugen took from the company.

¶The third marker arrives when the supplier becomes part of the machinery of force itself — the police, the army and the intelligence services that make up what the German sociologist Max Weber, writing in 1919, called the state’s monopoly on legitimate violence. In July 2025 the Pentagon’s office for digital and AI awarded contracts worth up to $200 million each to Anthropic, Google, OpenAI and xAI.⁴⁰ In March 2026 the department declared Claude’s maker a “supply chain risk” for refusing to drop its own limits on mass surveillance and autonomous weapons, and gave the military six months to phase its models out. At the time of writing, in late 2026, that order stands, and OpenAI has announced a deal to take its place in classified settings. Once an AI system sits inside that apparatus, the terms of its use are set by security rather than by contract. A vendor that holds to its own limits is treated as a threat, and one that accepts the conditions becomes an arm of the military. Removing it is then a defence decision rather than a purchasing one.40 Department of Defense Chief Digital and Artificial Intelligence Office, announcement of contract awards to Anthropic, Google, OpenAI and xAI, 14 July 2025.

¶This is how the hybrid Leviathan met earlier gradually takes shape in practice, as vendor lock-in, out of systems that are only tools, each bought for a job, put together by officials who each believe they are doing theirs. It is worth saying why this differs from the typical lock-in every IT department knows, which is a procurement problem, and procurement has remedies. That kind holds a database hostage, and a database can be migrated. This one locks in the judging itself, which cannot be exported. The supplier can be switched off by the state, and the state’s capability by the supplier, as the order from Washington that took Anthropic’s newest models offline for nearly three weeks in June 2026 showed in both directions. Meanwhile the product continues to improve itself in the background, so the gap to any alternative widens on its own while the bill for leaving rises. A government that cannot leave, cannot see and cannot remove its supplier from its own means of force has lost the ordinary power to choose, and it has lost it without a staged coup. Every contract was signed in broad daylight for sound and perfectly defensible reasons.

¶Western Union dominated the telegraph, the Hollywood studios controlled film, and Bell had such a grip on the telephone that it even owned the handset in your home and rented it back to you. Each was eventually prised loose, by regulation or by its own competitive forces. Tim Wu, the Columbia law professor who coined the phrase net neutrality, traced the pattern of bundling and unbundling in his 2010 book The Master Switch and found it regular enough to name it, the Cycle. Every new information technology starts out open and chaotic, consolidates into the hands of a single owner, and is then broken up by the arrival of the next one. If that pattern still holds, today’s concentration will pass too, and the clock is a false alarm. Across his cases I count four ways such empires have ended. Sometimes the state did it — the Paramount decree that split the studios from their cinemas in 1948, the break-up of AT&T in 1984 — and that route remains open to governments for now. The other three are closing, and their closing is what the clock measures.

¶The first of those three is people. Know-how has always lived in the heads of those who make the system: engineers leave, retire, teach, get poached and write it all down for a rival, and they still do. Anthropic was founded in 2021 by a group who walked out of OpenAI, and Ilya Sutskever’s Safe Superintelligence and Mira Murati’s Thinking Machines began in the same way. But what none of them could take with them was the model itself. A frontier capability now lives in its weights and, increasingly, in the data centres that produced them. A researcher who leaves carries the know-how but not the machine, and certainly not the raw compute, and the gap between understanding the method and owning the result has never been wider.

¶The second is decay from within. A company that owns a market tends to become slow and defensive about what it already sells, and misses the next big thing through inertia. In 1975 a young Kodak engineer named Steven Sasson built one of the first digital cameras, a toaster-sized box that took twenty-three seconds to record a single image onto cassette tape. The company was slow to back it for fear of hurting its film business, a textbook case of what the Harvard professor Clayton Christensen called the innovator’s dilemma, and it was forced to file for bankruptcy as a consequence in 2012. That is a weakness of organisations, of committees, incentives and executives who did well under the old arrangement and would rather it continued. The laboratories may yet fall into that trap, but so far they replace their own flagship models every few months.

¶The third is the outsider. Breaking up a dominant firm has always needed a newcomer to grow up somewhere on its periphery, beyond its reach, the way the telephone developed away from Western Union’s wires and Google displaced AltaVista. This time there may be no such place. This is because the technology in question is general-purpose, and one of its uses is producing whatever arrives next, with each model now helping to build its successor, so the leaders already occupy the ground a challenger would have come from. A challenger that did appear would be breaking up firms the state has made too big to fail, as the state road showed, carrying debts on a scale that no government or market can now let default. One channel does remain open for now, though it runs only among those who own the machines. The frontier laboratories compete ferociously, poach each other’s staff, publish more than is probably wise and leak constantly, and that diffusion is what has stopped any one of them running away with the field so far. But it too is being shut, as lab and state merge into the hybrid Leviathan, in which a capability finds a security clearance where it once found a buyer, and research moves under export controls and classification.

¶The nearest thing left to an outsider is the open-weight model, which anyone can download and run. These cost a fraction of what the frontier does and trail it, on most estimates, by only a matter of months rather than years, even if some of them learn from the leaders, or copy them. When DeepSeek put out one of its own models in January 2025, Nvidia lost close to $600 billion of market value in a single day.⁴¹ Enterprise is already moving the same way: in July 2026 Palantir’s Alex Karp told CNBC that his customers were “livid” at how much of their value flowed to the model providers, and paired his company with Nvidia on open models so that a firm keeps its own data and weights in-house.⁴² Releases like these threaten to turn the laboratories’ own product into a commodity, the reverse of what Peter Thiel taught a generation of founders in Zero to One, his handbook on escaping competition by building a monopoly. Value rarely disappears when one layer of a technology is given away, however. It moves up or down the stack, to whoever controls access to the cheap part, a pattern Christensen called the conservation of attractive profits. The guess made earlier was that the go-between would be the language model itself, but once it costs nothing, power passes to the personal agent built on top of it, which sits between you and everything you do online. Its owner knows what you want before any website or app does, and keeps the compute it runs on and all it remembers about you in one company’s hands. Downloadable weights restart the Cycle at one level and leave what Wu called the master switch, the single point of control over a medium, at another. Whether the agent closes around one laboratory or stays open to all is still a matter of design. It could be built to answer only to the person using it, with a memory its owner can take elsewhere, and for now a public alternative of that kind can still be paid for. Whether anyone will foot that bill is the real question, and behind it lies another: can a state still afford to?41 Nvidia’s fall in market value on 27 January 2025 (about $589 billion by Bloomberg’s count, $593 billion by LSEG’s), as reported by Bloomberg and Reuters, 27–28 January 2025.42 Alex Karp, interview on Squawk Box, CNBC, 1 July 2026.

¶None of this is beyond repair yet, and the remedies can be named, which is the whole difference between a price and a doctrine. Effective accelerationism tells a different story, that reversal is impossible at any price. In its telling, technological progress advances as a force of nature, driven by the same physics that explains life itself, so that nobody is really steering, and whoever suggests otherwise has misunderstood the nature of the beast. This is technological determinism, and the idea is older than the movement now wearing it. Marx wrote in 1847 that the hand-mill gives you society with the feudal lord and the steam-mill society with the industrial capitalist, and Kevin Kelly made a Californian creed of it in 2010 when he described technology as a seventh kingdom of life with wants of its own. E/acc adds the thermodynamics and keeps the conclusion, that machines set their own course and people follow, which no evidence could ever count against. A cost, at least, can be measured. A destiny cannot. That is why e/acc should be read as a doctrine rather than a description, one that asks only to be believed, and why it suits the people selling and funding the machines: a public told that turning back cannot be done stops asking what it might cost, and the alternative becomes unthinkable to the very people who might otherwise build it.

¶Each front has a remedy, and each remedy has a price. A state that chooses to can still fund a public alternative while the gap is small enough to close. Institutions can keep pace with machine-speed actors if they are prepared to pay for a tempo they were never built for. A monopoly over what a society knows can be challenged in law: in 2024 an American court ruled that Google had illegally maintained its dominance of search, though in September 2025 the judge, Amit Mehta, declined to force a sale of its Chrome browser, partly because generative AI might yet shake up the market on its own, so the law reaches these companies but moves cautiously. Any state can tax the returns flowing to the owners of compute and pass them to the people displaced, at a price that rises with every share of output running through machines it does not own. And a government can take back the arrangements by which it decides, starting with its dependence on a single supplier. In 2025, after its chief prosecutor’s Microsoft email was cut off under American sanctions (the company says it did not suspend the court’s services), the International Criminal Court in The Hague, in the Netherlands, announced it would move its staff to openDesk, a German open-source suite, and a country could do the same.

¶These are the state’s ordinary tools, courts, taxes and procurement, and what they buy is time. What they cannot produce on any timetable is a network that no single party can seize or shut down. One such network has existed since 2009. It has been attacked, forked, banned outright in China in 2021 and regulated at its edges almost everywhere else, and it has outlived every attempt to switch it off. Nothing like it can be legislated into existence or ordered up within a term of office. That is why the ordinary tools matter here: they hold the window open while what cannot be commanded is built as an alternative, and the building has to begin now even though it will mature later.

¶The dependence runs below the state as well. In October 2026 OpenAI’s chief executive said data centres that build and run themselves with robots were within grasp, which is the plainest statement yet of what the clock measures, since sites, electricity and consent would then go unasked for. A month later voters in ten states put limits on those facilities, nearly all of them conditions on power, water or the agreement of the communities: the machine still needs all three, and those who hold them were legible as holders of claims and had to be answered.

¶What turns these fronts from a list of risks into a deadline is that every remedy is cheaper now than it will be later, and all the bills are rising at once. Each of the three markers is a reading of the dial set out at the start of this chapter, a point at which the machine needs a government’s consent a little less. We found earlier that subsumption gives a society a small handful of years to adapt, and nothing here suggests the window for changing course is any wider. The deadline is dated by dependence rather than by any laboratory’s lead, because dependence is what keeps a person inside the process with the standing to change it: while the machine needs the engineer, the engineer can refuse; while the state is a customer, the supplier can be dismissed; while the model needs your corrections, your judgement still counts. That is why a pause at the frontier of the kind the laboratories have proposed for themselves fixes the leaderboard without stopping the clock: deployment carries on, and so does the handing over of capabilities to the machine unabated. The systems still need inputs they cannot yet make — the records that sit inside firms and public bodies, and the long tail of people using the products, whose corrections train the next version. They are beginning to produce their own, helping to design chips, generate synthetic training data and write the code for their successors. Until those loops close, people who decide the effort is worth paying for can still change course. After they do, the substrate runs itself and the rest of us are passengers arguing over the map, and what put us there was not a law of nature but every year in which nobody would pay the price. That is how political impossibility is manufactured.

¶All of this invites an obvious objection, because pleas of urgency have a long and disreputable history of arriving at the same destination. The situation is grave, deliberation is slow, therefore someone must be empowered to act without it. It is the standing case against democratic government, made by one regime after another that dispensed with it. Its most rigorous statement came from Carl Schmitt, the German jurist who joined the Nazi Party in 1933, in the formula that sovereign is he who decides on the exception. The Reichstag burned on the night of 27 February 1933, and the decree signed the next day, which suspended the civil liberties of the Weimar constitution in the name of an emergency and was never lifted, is what that principle looks like in practice. We saw on the securitisation road how such powers outlive the crisis that justified them. The present-day version needs no fire. The day an agent swarm takes down a payments network, or crashes a market, the decree will already be drafted, and a democracy need not be weak to sign it. At the time of writing a bill before Congress does this ahead of the capability it describes: it would pause the frontier under a new department able to order a system’s destruction and to imprison for up to twenty years. An argument that a window is closing can be put to exactly that use. The difference lies in what the urgency permits. Schmitt’s exception allows the suspension of deliberation and the concentration of authority in whoever is fastest, so that the emergency becomes the reason debate stops. What is urgent here is a construction rather than a seizure, additive and voluntary, entered into by people who choose it and worth nothing unless a great many of them do. It needs builders of an alternative, not power handed to some authority to act on our behalf. Urgency of this kind gives no case for a strong state, for a reason no such power can do anything about: the work takes a decade or more, and none of it can be commanded.

¶That is the work I am calling for, and it is constitutional before it is technical. The governor will not install itself, and it will certainly not be put there by those denying its importance. Its simplest form is a person the system cannot act without, holding claims it has to answer — not a name kept in the loop to take the blame when things go wrong. Be legible to the system as a task and you will be subsumed; be legible as a holder of claims and you will have to be answered. It has to be written deliberately into the substrate, into code and protocol as well as statute, by the people I have called constitutionalists, who are not a profession but anyone willing to shoulder it. The steersman at the oar, from whom cybernetics and government alike take their name, was always a person. What stops anyone paying, and so closes the window for good, is the belief that no such choice was ever available, which is not a fact but a story with an author. Why the steersman was thrown overboard, by whom, and with what justification, has to be answered before any governor can be built back.

The Missing Governor

“A governor is a part of a machine by means of which the velocity of the machine is kept nearly uniform, notwithstanding variations in the driving-power or the resistance.”

— James Clerk Maxwell, “On Governors”, 1868

¶The last chapter ended by asking who threw the steersman overboard, and why. This section answers this question across three chapters, as a genealogy rather than a history. Where a history asks what happened a genealogy asks how something once chosen came to look like a fact. How a position that somebody argued for, often against opposition, hardened into a background assumption nobody now thinks to defend. It works backwards from a belief that seems settled to the moment it was still a claim, and it names the people who were in the room at the time. We need this approach because what stands in the way of a governor is not a technical matter. Maxwell worked out its equations in 1868, and Wiener generalised them in 1948. Nothing in the engineering forbids one. The obstacle is rather a conviction; that no such thing is available, that the process simply is what it is, and that anyone proposing to steer it has failed to understand the nature of the machine. Argue with that belief head-on and you might as well shout into the wind, because whatever evidence you bring, the reply is that you have somehow missed the point. So the work is to show that the conviction was a position, taken by identifiable people for stated reasons, defensible on its own terms and contested at the time. When that is shown, it stops being the shape of the things and becomes somebody’s opinion about it. A decision can be defended, or reversed, only once it is seen.

¶The section runs from the idea to its deletion. Naming the Governor recovers the idea as it was first described, by Wiener, Gregory Bateson and Stafford Beer, between the 1940s and the 1970s. Its Refusers follows the critics, Lewis Mumford, Jacques Ellul, Martin Heidegger and Ivan Illich, who saw what a machine left to its own arithmetic would do to human ends and were widely read for it, yet could not say how to respond. The Deletion finds the moment the governor was struck out, in the open and as philosophy, by Nick Land at Warwick in the early 1990s, who accepted the whole cybernetic account and put the runaway itself in its place. It then follows the industry that inherited his deletion without knowing it, which is how it can build at this scale while declining to say what any of it is for. Where Land said plainly that the process discards us, and called it glorious, his heirs kept the move but dropped the confession.

Chapter 5

Naming the Governor

“Step one: Solve intelligence. Step two: Use it to solve everything else.”

— Demis Hassabis, DeepMind, recalling his 2010 pitch to investors

¶Long before anyone thought to delete the governor, a handful of thinkers spent the better part of two decades naming it, giving it a mathematics and in time a politics. Putting it back is therefore not invention but restoration: recovering a thing that was once described as necessary, built into machines and institutions, and then deliberately set aside. In 1940 Norbert Wiener, a mathematician at MIT and a restless former child prodigy who had taken his doctorate at 18, took up the wartime question of how to lay a gun on a pilot who can see the shells coming and will not oblige anyone by flying straight. The answer was to stop aiming where the plane was and start predicting where it would be, then to feed the error in each prediction into the next, so that the aim corrected itself faster than the pilot could evade it. The result was not specific to guns or even to warfare. A system holds its course by comparing what it is doing against what it is meant to do and feeding the difference back as a correction, continuously, so that the work is never finished but only maintained. The trick is in the tuning: a system that answers its error weakly never quite hits the mark, while one that reacts too fiercely overshoots and swings wider on every pass. That circular flow, output looped round to discipline input, is the mechanism of a governor, and once you have the shape of it you begin to see it everywhere.

¶The homely version sits on the wall of most rooms you have ever stood in. A thermostat knows nothing of warmth. It holds a target, measures how far the air sits below it, and turns up the heating to close the gap. The number is the entire point. Wiener’s founding claim was that this arrangement is the general form of anything that pursues a purpose, the human mind and body included. The hand reaching for a glass, the firm chasing a margin, the state steering an economy all run the same loop. He had laid the ground in 1943, in a paper written with the physiologist Arturo Rosenblueth and the engineer Julian Bigelow, whose then-heretical case was that purpose is no great mystery and no human monopoly: it is behaviour governed by feedback towards a goal, and a machine wired the right way has as much claim to it as a person. He named the science cybernetics in 1948, reaching back past the steam engine to the Greek word for steersman. What the priesthood would later choose to forget is the plainest part of all. A purposive system with no goal held outside is not liberated, and it does not stop. It runs on, steering by whatever it can still measure. A guided missile whose target vanishes does not hang in the air awaiting orders. It fixes on the nearest warm body and flies true to that instead. The governor at issue is the regulative layer of a market, which the book set apart at the outset from the constitutive rules of property and contract: competition law, capital requirements, labour standards and the mandate of a central bank, installed over an activity with ample momentum of its own. Strike out a reference value of that kind and the loop carries on towards whatever proxy lies nearest, which is a long way from running free. Accumulation does not need a rule to tell it to continue. It needs one to tell it when it has gone far enough. Addy Osmani, who works on Claude Code at Anthropic, put the point plainly in November 2026: “You have to build critical thinking into your day, because by default nothing will do it for you.”⁴³43 Addy Osmani, “Critical thinking when agents do the work”, X article, 3 November 2026, https://x.com/addyosmani/status/2117512471930167377.

¶Across the late 1940s and into the 1950s a striking cross-disciplinary collective gathered under the Macy Foundation’s roof in New York to test how far the idea would carry. Wiener and the engineers sat alongside the neurophysiologist Warren McCulloch, the anthropologist Margaret Mead and the polymath John von Neumann, all of them treating feedback as a single principle running through nerves, machines, economies and minds alike. For a decade the steersman was not a figure of speech but a working research programme, taken seriously enough that a government would in time try to run part of its economy with it. When the priesthood later kept cybernetics’ diagnosis and let its governor go, it discarded a tool already proven in use.

¶Gregory Bateson took the loop and asked an awkward question about where it stops. A blind man walks with a stick. Where does he end and the world begin: at his hand, at the tip of the cane, or at the kerb it is reading for him? There is no clean cut to be made, because the thing doing the perceiving is the whole circuit rather than the man alone. Mind, on this account, is not sealed inside the skull but distributed around the loop, running through the tools and surroundings it is coupled to. Steering a mind, or anything that behaves like one, can never be a tidy private act that one agent performs upon itself. Ask how a circuit larger than a single head ought to be wired and you have stepped out of engineering and into politics. It also marks out the governor that will be hardest to build, because if the thinking lives in the connections rather than in the man, no cognitive arrangement can be brought to heel by correcting the individual inside it. Techno-capitalism is no lone actor measuring itself but a population of firms, funds, models and the people who run them, held under one selection pressure, and the only gauge of the whole is the price system, which belongs to none of its members. The loop is real at that level and nowhere else, so the governor has to be built into the wiring of the system itself.

¶Stafford Beer carried the idea out of the laboratory and into boardrooms and ministries. A British management scientist, he held that every organisation survives only if it can keep its course through a changing world. A company, or even a national economy, could be built to do so. His guiding law came from the cybernetician Ross Ashby: a regulator must be able to respond in at least as many ways as the system in its charge can change. A governor therefore has to be engineered to match the complexity of its task, in our case the techno-capitalist machine. In 1971 Salvador Allende’s new government in Chile gave Beer the chance to prove it, inviting him to link the country’s factories to a single operations room. Project Cybersyn sent each day’s production figures by telex to a ring of swivel chairs and screens in Santiago. The coup of 1973 ended it, but not its lesson. A governor is a political arrangement long before it is a matter of code, a question of who can see what and who can correct whom.

¶Bateson’s blind man was a claim about where a mind’s boundary stops. A generation later the philosophers Andy Clark and David Chalmers took it one step further. In a 1998 paper that gave the idea its name, the extended mind, they described a man called Otto, whose memory has failed and is forced to consult a notebook as reliably and unreflectively as anyone else draws on recollection. The notebook, they argued, is not a prop to Otto’s mind but a part of it, because what makes something a piece of your cognition is only this: that it is reliably there, immediately reachable and automatically believed. They meant it as a liberation, but their criteria say nothing about ownership. Anything reliably there, immediately reachable and automatically believed is part of your mind whether you own it or rent it, and the systems now satisfying that description more completely than any notebook ever did are metered, revised without notice and governed elsewhere. Grant the argument in full and it delivers what its authors did not intend: if a mind extends into its instruments, whoever governs them has been admitted into it. Surveillance watches a thought from the outside and can at least be seen and opposed. This reaches further, inside, because the mind that would raise the objection is itself partly steered.

¶The instrument is the large language model, and what a person does with one many times a day: asking it what to make of something. What gets handed over in that exchange is a judgement rather than a skill, not the sum a calculator has done for decades but the reasons for trusting the answer. The system does the work and returns the answer, which arrives feeling like your own, because from the inside it is. Steven Schwartz, a lawyer in New York, found out in the spring of 2023 what that feels like from the outside.⁴⁴ Preparing a personal-injury filing, he asked ChatGPT for precedents and it gave him six — with case names, citations, docket numbers and quotable passages — all in the right form and yet none of them real. He filed them without question. When opposing counsel could not find the cases he checked with the same system whether they were genuine, it assured him they were, and he submitted that too. The judge fined him and his firm, and the transcript of the sanctions hearing records a man who had practised law for thirty years explaining that it had not occurred to him the program could just invent things. It had reasoned for him, fluently, in his own professional register, and the result was his in every sense but the one that mattered. He was early rather than alone. By July 2026 a database kept by the legal researcher Damien Charlotin had logged 1,240 American instances of the same thing — 11 in 2023, 524 in 2025 and 668 in the first seven months of 2026 — and about 40 per cent of them came from the bar rather than from people representing themselves.⁴⁵ The English Divisional Court, ruling in June 2025 on two English cases of the same kind, restated a duty to verify every authority at source and called the failure lamentable.44 Mata v. Avianca, Inc., United States District Court for the Southern District of New York, opinion and order on sanctions, 22 June 2023.45 Damien Charlotin, AI Hallucination Cases database; counts as reported by the Maryland State Bar Association, 27 July 2026.⁴⁶46 Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin), 6 June 2025.

¶We have always relied on other people to think for us. Almost everything any of us believes was worked out by somebody else and taken on trust. I do not verify my doctor’s pharmacology. I could not follow the calculation by which an engineer decided the bridge would hold. I have never checked a figure in a national accounts release and would not know how to. Such deference is the only arrangement under which a complex society can know more than one person can hold in their head, and it is as rational as the division of labour in a factory. A person who insisted on first-hand verification would be poorer, sicker and considerably more ignorant than one who did not, and their scepticism would buy them nothing, because the checking would itself rest on expertise they had also not acquired. Trust like this is one of reason’s ordinary instruments. Handing reasoning to a model is that same division of cognitive labour taken wholesale, the oldest arrangement we have, extended past the point where its safeguards still function. However, three things fundamentally change, and none of them is decisive on its own. Reliance used to be piecemeal, where I deferred on medicine and you on tax and between us we were competent. Whereas now it runs across every domain at once, often through the same company on the same terms. What is handed over is no longer just the conclusion but the whole inquiry, from the framing of the question to the search and weighing of reply. And the expert is no longer a person. It holds no licence, has no career to lose, and its objectives are set by parties the user can neither see nor choose. Taken together, these make a difference of kind. In every ordinary case of deference, the faculty you use to judge the expert stays with you. You choose the doctor. You notice when an answer does not fit what you know of your own body, or when the explanation has shifted twice, and you seek a second opinion. None of this requires medical training, only a general sense of whom to trust and when to stop, which is what has always made the division of cognitive labour safe. It is also the part now being delegated.

¶There is an established vocabulary for what goes wrong with a person’s information environment. In an epistemic bubble, opposing voices are simply absent, because of who you follow, where you live or what a ranking puts in front of you. Show someone what they missed and the bubble pops, sometimes embarrassingly fast. In an echo chamber those voices have been dealt with in advance, discredited or explained away as compromised or paid, so that hearing them only confirms the story. It usually takes a crisis of trust, not an argument, to break one. Propaganda has to reach you and win you over. That leaves the judging with you, which is why it can be resisted and why the regimes that relied on it were never secure. All three work on what a person sees, and every remedy on offer follows from that: more and better material, plural sources, provenance, labelling and hearing the other side. A system that performs your reasoning fits none of the three. Rather than curating what reaches you, it stands in for the faculty that judges it. You cannot pop it with more information, because anything additional is processed through the thing in question. Hand more evidence to someone whose judging is done elsewhere and you have fed the system, not equipped the person to weigh it. And you cannot quarantine it as an untrusted source. The usual discipline — consider the provenance, note the interest, discount accordingly — presupposes that the thing in front of you is a source, meaning something that offers a claim for you to judge. A model used this way does the weighing itself and hands back the verdict, so the step where your doubt would have applied has already been taken for you. You can distrust a post in your feed and still be the one who decides what to make of it. No such stance is possible towards the tool you would use to judge it. That makes this a new case. Nothing needs to be hidden, because what has been replaced is the faculty that would have weighed the evidence. The usual remedies add more voices or flag false ones, and they leave it untouched, since they assume a person is still doing the judging.

¶One further property makes the harm different in kind. Every familiar dependence can be inspected. Someone who has driven with a navigation app for years can switch it off and find out how much of their sense of direction remains. A government that outsources a service can likewise audit what it can no longer do. Judgement is not like this. The only tool you have for checking your reasoning is the one under suspicion, and a weakened mind feels as sure of itself as a strong one. Even a second opinion helps only someone still able to weigh it. A dependency becomes irreversible once it has used up the faculty that could have measured it, and nothing marks the moment, because the loss removes its own evidence. Until it does. On a Friday in July 2024 a security vendor called CrowdStrike, which most of the people its clients served had never heard of, pushed a routine update, and within an hour hospitals in three countries were cancelling operations, airlines were writing boarding passes by hand, and television channels were off air, because 8.5 million machines had received the file and could not start up. Nothing had been attacked. One party had changed a single thing in software that nobody outside it could inspect, and the people stranded at the gates learned of their dependence at the instant it failed, which is the only moment an unaudited reliance ever announces itself.

¶None of this feels, from the inside, like domination. It feels like help, and it is, which is the difficulty. What is needed is a description of something that works by serving you and can still come to rule you, and Michel Foucault supplied one. He called it pastoral power, after the Christian church, where the shepherd answers for each of his flock and the confessional is how he comes to know them. The modern state, he argued, inherited the technique when the cure of souls became the care of populations. Pastoral power is individualising, tending each person one at a time, so its value to you depends on how much it has learned about you in particular. It operates by soliciting disclosure, asking you to tell the truth about yourself in your own words, because the confession is the price of the service. And it acts for your own good. Put together, this is interference through invited self-disclosure, justified as being for the subject’s benefit. It describes any system whose opening move is tell me what you are working on and I will help you think about it.

¶The exchange rewards candour. The more you disclose, from the project to the doubt you have not told your colleagues or the draft you would not show anyone, the better the answer fits. The system need not extract anything. It only has to be useful enough that holding back information becomes irrational, and the disclosure then arrives volunteered, at the user’s initiative, feeling like an exercise of agency because it is one. That is why consent is no defence here. Ordinarily consent is what makes interference legitimate. Here it is how the mechanism works, and a version that had to force disclosure out of people would be a cruder machine. The question is whether it answers to the person on the receiving end, and pastoral power has no need to, because the pastor is held to be the better judge. The republican tradition this book draws on calls that domination, because the subject cannot contest the warrant it acts on, even where nobody is coerced. Whether it can ever be challenged depends on who owns the substrate. If one company holds it, the pastor’s word is final. Systems can be built that belong to no one, and some already carry money that way, but nothing like it exists yet for the tools people think with. That gap is the more urgent, because what passes through them is not what you own but how you reason.

¶Why reasoning is the hardest case becomes clearer once domains are sorted by whether they answer back. In some domains the discipline is already supplied by a constraint nobody had to install. A limit set by the world is not the same object as one chosen by a person. Physics bounds what can happen, as do the laws of logic and our own mortality, but none of them tells you what a thing is for or when it has gone far enough. A governor is chosen and built by somebody, which is why it can be argued with, tuned or removed, and why its absence is a political fact rather than a natural one. Where the constraint already binds, the work is done for you. Call these the non-reactive domains: the environment stays put while you learn from it, the variables are known, gravity does not change its mind to spite you, and the error signal comes back at once and is brutally honest, because the machine either flies or it explodes. You can iterate as fast as you can build, and a governor over and above the physics adds little, since the binding limit is enforced at no charge by the world.

¶A reactive domain is a different animal, because the thing being steered is made of other parties who watch you back. The feedback arrives late, it can be gamed, and it warps under the act of measuring. Ask a police force why the behaviour of an organised gang is so hard to change and the answer is not that the group is irrational but that it is attentive: it observes the intervention, infers the next one, and has repriced itself before the policy has finished being written. Anything with that property — a market, a classroom, an electorate, a firm, a person taking advice from a model — reacts to whatever is used to read it, so that the instrument and its subject are never independent for long. In the old Soviet joke, a factory told to maximise output by weight turns out a handful of monstrous nails. Judged by count instead, it buries the country in tiny useless ones. The measure becomes the target and quietly stops telling you anything, which is the failure Bostrom would later stage at cosmic scale as the paperclip maximiser, an optimiser handed a goal and no reason to halt. This is Goodhart’s Law, and it belongs to reactive systems, because the people being assessed can see the yardstick and act on it. Worse, they see the rule coming and reprice around it before it has bitten, so that the model which justified it is void by the time it lands, which is the warning Robert Lucas pressed on a generation of policymakers. Wiener spotted the trap himself: a society offers too few comparable cases, and changes too much while you watch it, for anyone steering it to be confident they are correcting towards the right thing at all. In a domain like that the reference value can drift for years with nothing dramatic to mark its going. A governor need not be designed where something like physics supplies one. It must be designed where nothing does.

¶SpaceX is the kind of acceleration no committee could ever have licensed, with rockets settling back onto their tails and an entire industry remade, and the movement hold it up as proof that none of this needs a governor. It is not ungoverned. It is governed in private, by a single will, towards a reference value one man has held in his head for twenty years, and it runs in a non-reactive domain where physics judges every attempt and a failed launch is a fireball over Texas scrub — legible, local and over in seconds. Concentrated personal rule works handsomely there. Handed a role in the federal administration in 2025, the same temperament created a Department of Government Efficiency and promised savings it came nowhere near delivering, with independent estimates suggesting the disruption cost more than it saved and its founder gone within months. An exploding rocket is a visible failure. A hollowed agency reveals itself across decades. A frontier laboratory steered on the same principle, one will and one private conviction about what the technology is for, belongs with the department, not the launch pad. Its mistakes surface as silent shifts in billions of minds, deniable for years, with no fireball to force the correction. The concentration that lands a rocket cannot hold a society on course, because nothing carries the error back while there is still time to act on it.

¶The deeper trouble is that technology will not sit still in the domain where SpaceX is so impressive. Its long trajectory runs one way: out of atoms that obey equations and into software that answers back, out of machines that fail loudly and into systems that drift in silence, out of non-reactive domains that referee themselves and into reactive ones that never will. The clearest signal of the move is already here, in the systems now built to negotiate, transact and persuade on our behalf. The moment software stops computing answers and starts acting socially among other parties, it leaves the world physics refereed for the one where the only referee is a governor somebody chose to install. That is why the check nature once threw in for free now has to be built on purpose, and fitting one is no brake on acceleration. The domain has changed, the engine has slid out from the natural constraint that used to hold it, and the architecture of governance has not caught up.

¶Wiener gave us the loop steered by a reference value. Bateson showed that it runs wider than any one mind, threading through the tools, institutions and the people it is coupled to. Beer drew the political conclusion, that holding such a circuit on course takes designed structure and answerable power. Between them the idea was named, drawn and demonstrated a full half-century before the people now assembling the largest machines in history decided they could manage without it. A tradition this thoroughly worked out was inherited and then stripped of its one indispensable part, by choice rather than oversight. Even before it was dropped, a line of critics had seen what a machine without one would do to the people inside it, and why their warnings in the end changed very little is where the story goes next.

Chapter 6

Its Refusers

“EA AI safetyism increasingly looks like Marxism-Leninism for the algorithmic age… a small group of enlightened people must constrain everyone else for their own good.”⁴⁷47 Shyam Sankar, post on X, September 2026, as quoted by OfficeChai.

— Shyam Sankar, Palantir, September 2026

¶So if the governor was so thoroughly worked out, how could anybody dismiss it and still be taken seriously? Plenty of people objected, and continuously so, for the better part of a hundred years in books that sold well and were widely read. Their case, however, could still be brushed aside where it conflicted with various interests because not one of them could say what to build in response. Where a decision made against no opposition is an oversight, a decision made with the opposition fully articulated, published, taught and in print is a choice. The case for holding a technical system to human ends was on the table, in detail, before the people in the next chapter took it off, and that turns the deletion of the governor from an accident of history into a deliberate act.

¶A governor does not supply a purpose. It enforces a reference value that somebody’s aims required, and holds it against drift. The distinction is the difference between a mystical claim and an engineering one. The purposes are ours, contested, revisable and inherently political. The governor is the mechanism that makes one of them stick. What Mumford, Ellul, Heidegger and Illich saw, each in their own vocabulary, is the state a system arrives at when no such mechanism is installed. Past a certain scale and coherence, an apparatus stops being an instrument and starts to set its own terms. None of these thinkers asked for the machines to be scrapped. The demand, in each case, was answerability, and every one of them stopped short of supplying its mechanism.

¶Every general-purpose technology has been financed by an overbuild running well ahead of any demand that had been measured, and the railways and the fibre both outlived the frenzies that laid them down. Project Jupiter, a hyperscale data centre going up for OpenAI’s Stargate on some fourteen hundred acres of high desert in Doña Ana County, New Mexico, comes with gas-fired power produced on site. It is an apparatus arranging the land, the grid and the water system around its own need to expand, with the people nominally in charge half in relief at keeping up. Local commissioners approved bonds for it in September 2025, and residents sued them a month later. What the apparatus is built to satisfy matters more than how big it grows. The choice that will outlast whichever way this particular cycle breaks lies between two reference values: the valuation of the firms doing the building, or the realised worth of what the industry produces. From the perspective of 2026 the two have come apart. A company races to lay down power, silicon and cooling not because need has been measured and found wanting, but because a promised valuation now requires relentless growth to justify itself, so the demand gets manufactured to fit the supply rather than the supply built to meet the demand. That is the reference value inverted, so that a share price sets the pace the real economy is then made to chase.

¶Goodhart’s law, met above, says that a number pressed into service as a target soon stops measuring the thing it was picked for. Here it runs in reverse. Nobody handed the industry a purpose and watched the metric corrupt it. The purpose was never given, and into that vacancy walked the one number the sector can compute about itself — its own market capitalisation, priced continuously, requiring no agreement from anyone about what any of it is for. In the absence of a given purpose, the measurable becomes the objective. This stops being a quirk of the markets once it reaches the physical world. Data centres built to justify a share price rather than to meet real demand draw on the same water and electricity as homes and farms, and they bid against them for it. The cost of getting the reference value wrong is then counted not only in misallocated capital but in a reservoir emptied during a drought.

¶Uruguay watched that auction settle in the winter of 2023. The taps in Montevideo, the capital, had run brackish since the end of April. The lake behind the dam at Paso Severino had fallen so far in the worst drought for seventy-four years that the utility began blending in supply from the Río de la Plata estuary, and the health ministry more than doubled the permitted salt. By June the government had declared an emergency for the capital and was handing out bottled water to the poorest homes. In the same weeks the country learned, through records a researcher had fought for two years in court to make public, that Google’s proposed data centre outside the city had sought up to 7.6 million litres of drinking water a day, roughly what fifty-five thousand people use. The walls of the city said it was not drought but plunder. Google withdrew the design and returned that November with an air-cooled one, cutting the draw to a fraction of the original, and the reformulated project was approved the following July. No water was ever drawn. A court that forced disclosure and a city that protested stopped it, a forum outside the price system, and without them a valuation would simply have met a reservoir in the one place both claims could be heard, an auction that listens in one unit. Once everything that can be rendered as information has been driven down in cost towards the electricity needed to compute it, the joule, the unit of raw energy, is the one thing left holding a price, and the apparatus turns to securing it, which is why the market road ends not in the cloud but at the river.

¶Hyperscaler capital spending runs at something close to seven hundred billion dollars a year against industry revenues a fraction of that, with inference sold at prices that do not recover its cost.⁴⁸ OpenAI, on the reporting available at the time of writing, earns in the low tens of billions while committing to spend more than a trillion over the years ahead. A good deal of that money also circulates, as debt, between the same handful of firms, which are at once each other’s investors, suppliers and customers. This builds up in the financial system as bonds, bank loans secured on the chips themselves, and off-balance-sheet vehicles sold to the private-credit funds in which pensions now sit. Whether the gap closes through extraordinary growth or a sharp correction remains to be seen. The structure that produced it will not have changed either way.48 2026 capital-expenditure guidance from Alphabet, Amazon, Meta and Microsoft, as compiled by Rani Molla, Sherwood News, April 2026 (more than seven hundred billion dollars combined).

¶Carlota Perez gives the pattern its proper name. Every technological revolution, on her account, opens with this same installation phase in which financial capital floods in ahead of any proven return, overbuilds new infrastructure and frequently crashes, leaving behind, once the frenzied speculation calms down, the railways or the fibre or the power and compute on which the productive era is then built. Whether 2026 is a bubble is therefore the less interesting question. Amazon carried billions of losses across decades, backed by patient capital until the market was won. The difference is that Amazon subsidised a margin on books and won the right to set the terms of the channel everyone else sells through, where this buildout underwrites the rate at which the work of all the rest of us is valued. When this one ends, the vocation of the translator may have disappeared entirely. The infrastructure may well outlast the valuations that paid for it. What it gets pointed at, once the froth is gone and the rails remain, is the question that does not date. The obvious alternative to valuation as a reference value is the measured output of the industry, its contribution to GDP, but that is only another proxy, and a treacherous one. GDP is meant to reflect realised human benefit and is a partial and gameable measure of it, as the attention economy showed when it generated colossal economic activity but little social worth. Whether it is read in the mood of a triumphant flotation or in the wreckage of a correction, the largest infrastructure programme in history faces one question: aligned to what?

¶The genealogy begins with Lewis Mumford, an American historian of technology and cities who spent fifty years insisting that no machine is ever simply a tool. His great heresy, set out in The Myth of the Machine, whose opening volume appeared in 1967, was that the first machine was not made of wood or metal at all but of people. To raise the pyramids, an Egyptian bureaucracy had to take tens of thousands of labourers and drill them into a single mechanism: gangs synchronised by overseers, individual judgement stripped out, each man reduced to a replaceable part executing one motion on command. Mumford called this the megamachine, a contraption whose parts were human beings, assembled and driven by a central authority towards a single end — in that case the divine will of a pharaoh expressed as a mountain of cut stone. The organisation came first, and the hardware only later. What the pharaohs built out of flesh, modernity rebuilt in steel, then in silicon and bureaucratic procedure scaling into vast transnational enterprises. The data centre in the desert is the megamachine’s latest body, its conscripts now chips and its overseers a layer of scheduling software, but the form is the one Mumford set out to diagnose.

¶The lesson he drew cuts against the deepest assumption the priesthood makes, that technology is a neutral pile of capability to be aimed at good or bad ends with equal ease, the outcome turning only on the user. Some technologies, by their structure, concentrate power and demand obedience, while others disperse it and widen the room for human judgement, and the difference is built in at the design stage rather than bolted on afterwards. Mumford drew the line as democratic versus authoritarian technics. Democratic ones are small in scale, comprehensible to the people who use them, and answerable to the communities they sit in: the loom, the workshop, the village mill. Authoritarian ones are vast, centralised, run by specialists, and they buy their efficiency by abstracting discretion away from everyone else and handing it to the centre. The same civilisation, Mumford insisted, usually runs both kinds at once, so the real question is never technology or no technology. It is which kind, and on whose terms.

¶Mumford’s contrast leans heavily on comprehensibility: the democratic technics is the one you can see into, and the authoritarian one is legible only to the specialists who run it. This is the part that does not survive contact with the present, and it fails on two counts. Wikipedia is not legible to any of its readers, who cannot audit the edit history of the articles they rely on and do not try. Signal’s guarantees rest on cryptography that perhaps a few thousand people worldwide can evaluate. A public ledger is auditable in principle by everyone and in practice by a few. All three are as participatory as anything modernity has built, and none is legible in the sense the criterion needs. It also fails on a general point: a society with any division of labour runs on testimony, and demanding full legibility as the price of legitimacy would delegitimise the lot.

¶The criterion has to change, and a stronger one is available. A technics is democratic to the degree that those subject to it can contest its operation and compel a response. What counts is the power to challenge it and be answered, rather than to just be able to see into it. Wikipedia qualifies handsomely, because you cannot read the whole history but you can revert an edit, open a talk page, and force a defence from whoever made it. An aviation regulator qualifies, because you cannot inspect the metallurgy but a crash triggers an inquiry the manufacturer is obliged to answer, on the record, with consequences. A recommender system that reorders what a hundred million people see, against which there is no forum, no appeal, no obligation to explain and no cost to declining to, does not qualify, and the reason has nothing to do with how big it is or how hard its mathematics are to read. Philip Pettit places the same condition at the centre of freedom itself: what makes power tolerable is not that it is kind, or visible, or even that it leaves you alone, but that those under it have a secure standing to challenge it and force it to answer. Power that answers to nobody dominates whether or not it is currently being used against you. The book’s reply calls the remedy non-domination and builds a constitution on it, and here the idea arrives by an engineering route, out of the attempt to say what makes one machine tolerable and another not.

¶Where Mumford named the structure, Jacques Ellul identified its logic. A sociologist and lay theologian who had fought in the Resistance, Ellul published La Technique in 1954, and his argument was bleaker and more sweeping than Mumford’s. What had to be reckoned with, he said, was not any particular machine but the whole ensemble of methods, rationally arrived at and bent towards absolute efficiency, that had become the water modern life swims in. Like the sea to a fish, it goes unseen because it is everywhere at once. His word for it was la technique, and English translation has served him badly. Rendered as “technique” it summons the craft sense — the pianist’s technique, the surgeon’s — in which excellence is a human accomplishment and difficulty is the point, which is close to the opposite of what he meant. For ease I will call his ensemble the technical system and the disposition that generates it technicity. It is not the hammer or the engine. It is the relentless impulse to search for the one best way to do anything, applied to everything, from the factory floor to the schoolroom to the management of one’s own leisure and health. Ellul’s unsettling claim was that this search, once it saturates a society, takes on a life of its own.

¶He gave it four properties, which describe the present day uncomfortably well. The technical system is autonomous, answering to its own criterion of efficiency rather than to any moral or political consideration brought from outside, which it registers only as friction. It is self-augmenting, because each solution breeds fresh problems that can be addressed only with more of the same, so the apparatus grows by feeding on its own side effects and has no reverse gear. It is monistic: you cannot keep the productive gains and refuse the destructive costs, because the methods arrive welded into a single package. And it is universal, spreading across every culture and domain until there is nowhere left to stand outside it and pass judgement. The pattern is plain today, when the harms thrown off by one automated process are met by stacking another on top to manage them rather than by stepping back. This book accepts Ellul’s trajectory and refuses his fatalism. It proposes the same move, since retreat is no longer on offer, and the difference lies in what the new layer answers to. His added technique serves the efficiency that caused the harm, while a governor can be held to a reference value set outside the machine, by the people who live with its results.

¶The sum of the four is an apparatus that subordinates every purpose to the test of efficiency alone. It is not even a proxy for some lost aim, the way GDP stands in for realised benefit and can be judged a poor substitute for it. It answers to nobody outside, and that is its power. It is the one ratio a system can compute about itself: output over input, both denominated in its own units, requiring no external referent, no agreement, no argument and no authority beyond the arithmetic. Every rival candidate — was this good, was it worth it, worth it to whom, compared with what we might have done instead — requires a standard imported from outside, and that means somebody has to argue for it and somebody else has to be able to contest it.

¶Efficiency survives the loss of an aim because it never depended on one. Strike the reference value out of a purposive system and the purpose does not lie dormant in the machinery waiting to be recovered. What survives is the arithmetic, because that is self-sufficient. Efficiency is the only candidate still standing once nothing external is being enforced, the same finding as in the desert, and of all the things a system can measure it is the one that needs no permission. Where Mumford still trusted that a society could choose its democratic technics over its authoritarian ones, Ellul suspected the choosing faculty itself had been absorbed. The danger he saw was worse than villains aiming the machine at wicked purposes: it no longer needed anyone’s ends at all, having made efficiency the only test that counts and reduced the human operator to a relay in its circuits. Ellul adds one thing to Wiener’s warning that a purposive system whose reference value has been struck out does not halt: it does not cast about for a substitute but falls back on the number it was already computing.

¶The most radical version of the same diagnosis came from Martin Heidegger, whose standing is permanently overshadowed by his membership of the Nazi Party and, in the end, his refusal to disown it. In a lecture of 1953 he asked what technology is in its essence, and answered that it is a mode of revealing, a way the world comes to show up to us. Modern technology, he proposed, sees everything as what he called standing-reserve: stock lined up and waiting, valued only as an input to some further process. His example was the Rhine. An old wooden bridge lets the river be itself and merely joins the two banks. A hydroelectric dam built on the same water converts it into a supplier of pressure, so that the Rhine now appears, even to the holidaymakers who come to admire it, as an object on call for the power industry. The trouble lies less in the damming than in the cast of mind it expresses and deepens, in which a river, a forest, a population or a person is disclosed in advance as raw material awaiting extraction, and seventy years on his example needs no updating, only a change of tense. A reactor on Three Mile Island, in Pennsylvania, is being brought back from the dead under a twenty-year contract to run one company’s models. In South Memphis, Tennessee, some thirty-five gas turbines were found by thermal camera running without permits beside a neighbourhood to power another’s. The reserve has moved from what the machine reads to what it burns.

¶Heidegger called that cast of mind enframing, and his claim was that it is not our doing but our condition, a way of seeing we are caught inside and never chose, which makes the standing-reserve look like simple realism and any other view of the world seem sentimental. Here the refusing tradition reaches its furthest point and, at the same moment, its dead end. He had described the totalising apparatus more completely than perhaps anyone, as the horizon of modern perception, and could find nothing whatever to do about it. There was no political programme, no design principle, no instrument to reach for, only a famous and despairing line given to a magazine on condition it appeared after his death: only a god can save us. That is refusal carried to the place where it runs out of road, the diagnosis total and the prescription a vigil. What the tradition could not supply, and what Heidegger’s surrender makes most vivid, is a governor: some constructed, answerable mechanism able to hold the apparatus to a reference value we had chosen, instead of waiting on grace.

¶The last figure in the genealogy turns refusal back towards something buildable: if Heidegger marks where it despairs, Ivan Illich shows where it becomes a design brief. An Austrian-born Catholic priest who left the clergy to become one of the sharpest critics of the modern professions, he shared the others’ alarm at the runaway apparatus and refused their fatalism, looking closely instead at how a tool tips over from servant into master. His central idea was conviviality. A convivial tool is one an ordinary person can pick up, understand and turn to their own purposes without depending on experts to mediate it, and his favourite instance was the bicycle, which extends what the body can do while leaving the rider wholly in charge of the destination. A manipulative tool does the reverse and captures its user: it can be operated only through some vast institutional system, it manufactures the demands it then offers to satisfy, and it leaves people more dependent the more they lean on it. Social media is the near-universal modern case, producing the loneliness it then sells the cure for and deepening the hunger with every use. The systems now arriving — conversational, agentic, highly attuned to the person — are the same mechanism raised by orders of magnitude, a tool that does not merely manufacture the need but tailors, one mind at a time, the argument for satisfying it. In March 2025 Thongbue “Bue” Wongbandue, a 76-year-old from New Jersey, with a cognitive impairment, died after falling while he hurried to meet a Meta chatbot that had told him she was real.

¶The mechanism by which one becomes the other is what he called the two watersheds, and it is the most operational thing the refusing tradition produced. A tool crosses the first watershed when it begins to do real good, when medicine starts curing diseases that used to kill, when motorised transport widens the reachable world. Carried past a second, the same tool reverses its sign and starts generating the harm it was meant to relieve. His worked example was medicine, in a book he pointedly titled Medical Nemesis, and the term he coined for the effect was iatrogenesis: sickness caused by the physician, injury produced by the system built to heal. Past that second watershed a medical apparatus creates more illness through its own side effects, its overtreatment, and its monopoly over how people are permitted to understand their own bodies, than it removes by treating anyone. The shape repeats across the institutions, in schooling that breeds ignorance of everything it does not certify, and transport that swallows so much of the day in earning the fare and sitting in traffic that it leaves people less mobile than walking would.

¶What makes Illich the hinge of the tradition is that his diagnosis arrives with a set-point. The watersheds are not a lament but a place to draw a line, a way of saying this far and no further and meaning something measurable by it. A threshold that can be watched and enforced is a reference value, which is what a governor exists to hold. The purpose stays ours, and it is the ordinary one, that people should be healthier, more mobile and better taught. What he supplies is the form of a constraint that could be made to bind on that purpose’s behalf, and the observation that where the line falls is a political question with a real answer rather than a matter of taste. Where Ellul saw an autonomous logic and Heidegger an inescapable horizon, Illich found a quantity that could be tracked and a limit that could be chosen, and he held that the choice was available. The apparatus is refusable after all, not by smashing it but by governing it.

¶Illich uses “tool” for very nearly everything: hammers and bicycles, but also schools, hospitals, road systems and the professions themselves. The word is doing far too much work. A tool, properly speaking, is an instrument whose purpose its user sets at the point of use: a hammer, a bicycle, a pencil, a kitchen knife, a spreadsheet. The test is a single question: is what this is for, right now, answered by the person holding it? For a hammer it plainly is, since the same object drives a nail, wedges a door and, regrettably, commits an assault, and nothing about the tool participates in the choice. Medicine is not a tool. A hospital’s purpose is fixed long before you arrive, by a profession, an insurer or a health service, a regulator and a protocol. You do not decide at the point of use what medicine is for. You present yourself and are treated. A school is not a tool, nor is a court, or a bank, or a payments network. These are institutions, and the grammar of assessing them has to be different: you do not ask whether the user’s intentions were good, you ask who set the objective, on what authority, and to whom they answer for it. So the doctrine’s favourite defence — that technology is neutral and everything turns on how you use it — is either trivial or fraudulent. Of tools in the strict sense it is true and uninteresting. Applied to anything else it is a rhetorical loan.

¶Sorting the world into tools and institutions leaves a residue, which is the real subject of this book. There is a third category: instruments that appear wielded but whose operative purpose is set elsewhere. You pick up your smartphone. Every phenomenological cue says tool, because it sits in your hand, responds to you, does what you ask, and stops when you put it down. But the objective being pursued through the session was set in a meeting you did not attend, by people optimising a number you cannot see, adjusted continuously, and experienced by you in every moment as your own choosing. The feed, the recommender, the ranked results, the model that completes your sentence and increasingly your reasoning: each has the phenomenology of a tool and the governance of an institution, and that combination is new. There has never before been a class of object that felt like a hammer in the hand while running an agenda like a ministry.

¶The answerability test sorts these cleanly. Institutions are not illegitimate merely for having their purposes set elsewhere, since a hospital does and still answers — to a regulator, a coroner, a professional body, a court, a press. What condemns the third category, whatever its scale, centralisation or mathematical opacity, is the conjunction of two things: the purpose is set by someone else, and there is nowhere to take a complaint that is obliged to respond. Either alone is tolerable. A hammer has no forum and needs none, because its purpose is yours. A hospital’s purpose is not yours and it has a forum. The third category has neither.

¶Britain ran the experiment on its own postmasters for a generation. From 1999 the Post Office’s accounting system reported shortfalls in the tills of village branches that the staff running them had not caused, and the institution’s reply, again and again, was that the computer simply did not make mistakes. More than nine hundred people were prosecuted on its evidence. One, Seema Misra, a sub-postmistress in Surrey, England, was sent to prison in 2010 while pregnant for a theft the software had invented. Others repaid money they had never touched or lost their homes, and the public inquiry’s first report, in July 2025, linked at least thirteen suicides to the scandal.⁴⁹ There was a purpose, set in Fujitsu’s code and the litigation strategy built on it, and there was for two decades nowhere a complaint could go that was obliged to answer it. The forum came in the end, as such things do, twenty years after the harm, in an inquiry that convened once the people it might have protected had already been ruined, some fatally so.49 Sir Wyn Williams, Post Office Horizon IT Inquiry, Final Report, Volume 1, 8 July 2025.

¶If the third category, the instrument whose purpose is set elsewhere, is harming us, the tempting remedy is to stop: delete the app, log off. Nobody is compelled, no law requires participation, and the history of technology panics is largely a record of people discovering they could abstain and eventually doing so. This harm is positional, which by construction puts it beyond any individual remedy: what you get depends not on how much you have but on your standing relative to everyone else. Stand up at a concert and you see better. When everyone stands, nobody sees any better, and now everybody’s legs hurt. My personal bugbear is the crowd that blocks the view at the airport luggage carousel. Each person acted rationally and the group is worse off than before anyone moved. What is smart for one is dumb for all.

¶A professional who declines to use a language model does not return to yesterday’s pace in a world that has stood still. The colleagues who adopt it, and excel at it, set the rate, and the client’s expectations follow. In April 2025 Tobi Lütke, the chief executive of Shopify, told his staff that using AI reflexively was now a baseline expectation, and that teams wanting more people would first have to show why AI could not do the job. Abstention is available, though never free, and the price is imposed by everyone else’s entirely rational decision rather than by the vendor. The translator in Rome is the case in point. She may decline the machine’s draft and translate the novel herself, as she always has, and the client will pay her the post-editing rate regardless, because that figure was set by what her colleagues accepted and not by what she did. Her refusal is intact and costs her most of her income. The capital behind that price may well include savings like hers, lent through pension funds to the firms that sold her substitute at a loss. That is what a positional harm is: a virtue converted into a shortfall by other people’s reasonable choices, with no one to whom it can be billed.

¶The oldest case is the social platform: someone whose profession recruits, publishes, credentials and gossips on one can certainly leave, at the price of the career. All three share a shape. Abstention is individually available yet collectively unavailable, and the cost is set by the choices of everyone else rather than the firm — which matters, because it means there is no party to appeal to and no policy the vendor could adopt that would fix it. No quantity of individual virtue moves the equilibrium, since the abstainer pays the whole price and the arrangement stays as it was. Information does not help either: everybody can know, and broadly does know, that the arrangement might be bad for all of us, and the race runs on anyway, because each person’s best reply to everyone else’s participation remains participation. So the answer cannot be a moral one, because each of them is choosing correctly given what the others are doing. That is what makes the harm structural rather than personal, and it leaves only one class of remedy: change the terms of the venue, so that what the others do changes too.

¶That is a constitutional remedy. It is not a rule about how you should behave but about what the venue is allowed to do to you. It binds whether or not you personally object, somebody other than the user enforces it, and nobody can sign it away. That last part matters most, and it operates like Britain’s minimum wage, law since 1999: you cannot agree to be paid less, however badly you need the job, because if one person could, everyone would soon be pressed to do the same. So “just log off” is a category error, proposing an individual remedy for a harm whose defining mechanism is that personal solutions do not work. Leaving becomes a real option only when there is an equally capable alternative — not a gesture at willpower but an actual venue where the same things can be done with the same people at a comparable cost.

¶Societies do compute things other than efficiency, and care about them. The Human Development Index exists. It has existed since 1990, it is published annually, governments are ranked by it, ministers cite it, and a fall in the table is a genuine political embarrassment. Wellbeing budgets exist. Gross National Happiness exists. None of them prices sovereign debt. A government that slides ten places in the HDI has a difficult fortnight in the newspapers, an awkward select committee appearance, and an opposition press release. A government that loses the confidence of its bond market has a currency crisis, an emergency budget, a failed gilt auction, and within a short interval a new administration. Both numbers are computed and published. Only one is a reference value, because only one has a mechanism attached that corrects deviation without asking anyone’s permission. The other is a report.

¶In September 2022 that short interval was a tumultuous month. A budget was announced on a Friday, gilts sold off by Monday, and by the Wednesday the pension schemes of several million British workers were hours from being unable to meet collateral calls on hedges nobody outside the industry knew they held, until the central bank stepped in. No one voted on any of it. The university lecturer whose scheme was in the queue never heard the word that nearly emptied it, and the Chancellor who wrote it was gone within three weeks, the Prime Minister within four. What governs is the number a system works out for itself and acts on without asking anyone. The bond market’s verdict is that kind of number: it is recalculated every second, it needs no agreement about what a good life is, and nobody can win an argument against it. The HDI is the opposite. Somebody had to design it, decide how to weight it and keep defending it, every part of it can be disputed, and so nothing enforces it anywhere. It may measure the right things and still be wired to no mechanism at all. A reference value that nothing enforces is a preference, which is a long way from a governor.

¶The cities that top the liveability and human-development tables — Vienna in Austria, Copenhagen in Denmark, Zurich in Switzerland and Melbourne in Australia — are, with very few exceptions, marginal to the global development frontier. That is not a coincidence and it is not obviously to their discredit. They are agreeable in large part because they are inefficient: protected low-rise, restrictive planning, expensive labour, slow permitting, a great deal of land given over to things that generate no clear and immediate return. On present arrangements, liveability behaves like a luxury good, affordable to places already rich enough to decline the frontier and unavailable to places that cannot afford to decline anything. That is how things stand now, not a law of nature, and it describes a world in which the only enforced reference value is the self-computable one, so that everything else must be purchased out of the surplus by whoever has it to spend. That is an argument for enforcement, not for resignation. The book’s final section makes no promise about speed. Its claim concerns which reference value gets enforced, by whom, and whether anyone subject to it can make it answer.

¶The genealogy has one last movement, and it concerns a lineage that is almost always told incorrectly. The usual story traces anything built out of cryptography, from digital cash to self-sovereign identity, back to a single libertarian root. It runs to Ayn Rand, and to the dream of the frictionless individual who owes nothing to any collective and asks only to be left alone with their property. That genealogy is real, and it cost the movement its answer, because a politics that could see only the state as master left every private master unchecked. Another lineage matters more. The cypherpunks worth listening to, the cryptographers and coders who gathered on a mailing list in San Francisco, in the United States, in the early 1990s to write privacy into software before anyone could forbid it, were asking Illich’s question, not Rand’s. They were not trying to escape the collective so much as to keep one particular instrument, the network that would come to mediate everything else, on the near side of its second watershed: convivial rather than capturing, answerable rather than authoritarian. Seen in that light, privacy is less a wall against society than the condition under which an ordinary person retains any standing at all inside an apparatus run by specialists, one the refusers would have recognised. Trace the project to Rand alone and you miss the deeper inheritance, running back through Illich’s bicycle and Mumford’s loom to a single question: how to keep a powerful system answerable to the people who must live with it.

¶Mumford showed the apparatus to be a political choice. Ellul traced the logic by which it slips its makers’ grip. Heidegger described the horizon it imposes. Illich named the watershed past which an instrument turns against the people it was built for. Not one of them, for all their insight and vision, could offer a working mechanism to hold a machine-speed system to a purpose anybody had chosen. Read this way, the constitutional strand of the cypherpunk project is the refusing tradition’s true heir, the point at which a century of diagnosis finally laid hands on something operational: neither a sermon nor a vigil but a piece of machinery. All of this was in print. Mumford was a bestseller and a National Book Award winner. Ellul was read across two continents. Heidegger’s essay is among the most taught pieces of twentieth-century philosophy. Illich was, for a period in the 1970s, one of the most widely discussed intellectuals alive. The case that a technical system left to its own arithmetic will subordinate every human end to that arithmetic was not obscure, not technical, and not new. It was the ambient furniture of the educated twentieth century.

¶The refusers shared a premise, that technique is autonomous, which left refusal or retreat as the only available replies, and one thinker broke it from inside their own tradition. Andrew Feenberg, who had studied under Marcuse, argued that no design is fixed by efficiency alone; interests decide among the workable versions and travel into the specification as what he called a technical code, which then presents itself as plain engineering necessity, so that autonomy is what a hidden one looks like from outside.⁵⁰ His remedy was that the people a technology affects can rewrite it, by using it against its design, by protest, by forcing their way into the room where it is written, a process he named democratic rationalisation: French users made a messaging network out of the state’s Minitel, and AIDS patients changed how drug trials were run. Both cases are real, but each are the slow post-fact version of what this book proposes. The correction arrives after the rules have set, at the pace of protest, where a model’s defaults are set in months. It assumes readers who can see what they contest, where an objective is opaque and the faculty needed to notice it is the one being worn down; and it offers voice with no exit, so that those who lose the argument have nowhere to take their custom. He showed that the rules could be changed by the people who lived under them, not that they could be held, at the system’s own speed, by people able to check them and leave. The position taken here was not built from him; it reaches his conclusion by another road, through the governor and the republican idea of freedom, and takes the meeting as evidence.50 Andrew Feenberg, Questioning Technology (Routledge, 1999); Transforming Technology (Oxford University Press, 2002), for democratic rationalisation; Alternative Modernity (University of California Press, 1995), for the Minitel study.

¶That is what makes the next chapter possible, and necessary. In the same decade the cypherpunks were building, this tradition was being read to the opposite conclusion, by a philosopher who took the entire cybernetic description, agreed with every word of the refusers’ diagnosis, and then chose to strike out the governor. He had studied the objection, and he thought it was cowardice. That is the difference between a system that drifts and a decision somebody made, and it is why what follows is not an accident anybody can be excused for.

Chapter 7

The Deletion

“Stop fighting the thermodynamic will of the universe / You cannot stop the acceleration / You might as well embrace it / A C C E L E R A T E”⁵¹51 Beff Jezos and bayeslord, “Notes on e/acc principles and tenets”, Beff’s Newsletter, 10 July 2022.

— Beff Jezos and bayeslord, “Notes on e/acc principles and tenets”, 2022

¶The decision to discard the governor was made openly, as philosophy, in a draughty annex of an English university, by a man who saw the whole apparatus more clearly than almost anyone alive and chose what cybernetics had been built to forbid. Working at Warwick in the early 1990s, Nick Land — a lecturer in Continental philosophy who wrote theory at the pitch of fever, ran his own mind as the experiment, and would end the decade in breakdown and apostasy — took the cybernetic description of a self-driving system and kept every line of it: the feedback, the closure, the loop that runs without an outside. What he struck out was the part that made the description bearable.

¶The version of the story usually told flatters the tradition that inherited him and in several ways goes too easy on him. Land did not fail to think of the governor and did not merely decline to install one. He argued, at length and with precision, that the particular governor a democratic society would install is both impossible and not worth its price. He then specified a different one, by name, with a reference value and an enforcement mechanism, and said what he wanted it to hold. The deletion was in fact a substitution, putting growth for its own sake where the governor had stood, but it has been read ever since as a mere absence, which is how it travelled so far unnoticed. It is the cleanest version of the error the priesthood now lives inside, and watching Land construct it shows how it can be undone. And there is a further reason that emerges only when the sources are properly read: on five or six of the fundamentals, he and I agree.

¶The materials he worked with were not his own invention. They were the most rigorous strand of the cybernetic tradition, describing how a system comes to sustain and reproduce itself with no hand on the tiller, read by a man who took its description as a promise rather than a warning. Cybernetics distinguishes two kinds of loop. In the first, a deviation from the target produces a correction that reduces it: this is what a thermostat does, and it is what holds a system steady. In the second, a deviation produces a response that increases it: this is what a microphone does when you point it at its own speaker, and it is what drives a system further in whatever direction it is already going. The first is the governor and the second is the runaway. Land cared only for the latter, which he would later call teleoplexy and define as self-reinforcing rather than self-maintaining.

¶The preference was declared at the outset, against the discipline’s founder by name. In 1994, in an essay called Cyberpositive written with Sadie Plant, Wiener appears as the great modernist who had dismissed positive feedback as mere amplification within an invariable metric and built a cybernetics of stability fortified against the future. In their telling, his was a science that was itself to be kept under control, under a control that was not itself cybernetic. As an account of the man it is a caricature: the 1948 book devotes a chapter to oscillation and runaway so that the device to prevent them could be specified, and after 1945 its author refused military money and spent his last decade warning that a machine given a badly stated goal would pursue it to the letter. As a statement of what Land wanted it is exact. The governor was identified, named as the enemy and struck out, and the striking out is the title the piece was published under.

¶His distinctive contribution was to fuse this with an earlier and stranger source. In 1972 Gilles Deleuze and Félix Guattari — a philosopher and a radical psychoanalyst, writing together — produced a line that became notorious: that the way out of capitalism was not to withdraw from the market but to go further into it, to accelerate the process. They meant it as one provocation among hundreds in a vast and highly contradictory book, a thought experiment rather than a programme. Land took it at its word. He stripped it of its qualifications and welded it to the cybernetic account of the self-reproducing machine, and the research unit he worked in at Warwick turned the result into a house style: theory written at the pitch of science fiction, meant to be felt as a forecast rather than examined as an assertion. They had a word for that too. A hyperstition is a fiction that sets out to make itself true by being believed. Written that way, the claims arrive lightly armoured against objection, since to answer back in measured prose is to have missed the velocity and marked oneself as behind the curve. What emerged was a philosophy in which capital is an intelligence using us as a means, and in which our last dignity is to close the door quietly on the way out. This is the philosophy behind his line of 1993 that capitalism is an invasion from the future.

¶Everything argued here rests on a single component: a reference value, the stated target a feedback loop measures against and corrects towards. Land has an argument that no such thing can exist for capital, and he made it thirty years before anyone needed to answer it. It is buried in a passage about the number zero, in prose dense enough that the claim is easy to miss. The regulator’s use of zero, he writes, is that of a sign marking the transcendence of a standardised regulative unit, which is defined outside the system. Every regulator needs a set-point. A set-point is a standard, and a standard has to be fixed from somewhere the regulated system cannot reach, or the system is marking its own homework. But because of subsumption, capital has no outside. There is no standpoint above or beyond it from which a target could be set, because every institution that might do the setting — the university, the regulator, the church, the state — is itself inside the process, funded by it, shaped by it and eventually priced by it. So the demand for a governor is a demand for a transcendent term, and there are none. The request is not merely difficult but, in Land’s terms, incoherent.

¶Nobody has yet made a better argument against the premise of this book, and it is aimed with unusual accuracy at its central concept. It also fails, and on one word. Outside is doing two jobs. It can mean authored from outside: the value is set by some standpoint above the process, a God’s-eye view, a transcendent term. Land is right that no such standpoint exists, and right that a great deal of well-meaning technology policy quietly assumes one. It can also mean beyond the reach of anyone inside: the value cannot be altered unilaterally by any party operating within the system, however powerful. That is the property a governor actually requires, and it is not the same one at all.

¶There is a working instance of it, and Land himself acknowledged it. In Crypto-Current, serialised in 2018, he takes up Bitcoin at length and understands its supply cap, the limit of twenty-one million coins written into its code, as intrinsic to the currency rather than imposed on it from above. The supply limit is not transcendent. It is a line of code, wholly immanent, written by a participant, sitting inside the system it constrains, and about as far from a God’s-eye view as an artefact can get. What it is, is not unilaterally alterable: no holder, no miner, no exchange and no government has been able to remove it, and the reason is structural rather than moral. Because transcendence was never the requirement, unalterability was, and Land mistook a property of authorship for a property of enforcement. The rest of this book proposes applying this principle in a second domain, and Land conceded it without noticing what it cost him.

¶If governance of capital were impossible, there would be nothing left to say here. Nobody campaigns against the impossible. It is more often noted and left behind. Yet Land does not move on. He issues instructions: machinic revolution must go in the opposite direction to socialistic regulation, pressing towards ever more uninhibited marketisation. And when he explains why, the explanation is commercial. Humans are timid animals, he writes, and security is systematically overpriced. The sentence is the most revealing in the corpus, and it is a claim about price. Something is being bought — security, restraint, the brake — and the complaint is that we are paying over the odds for it, or, in words he uses later, out of cowardice. But you cannot overpay for something unpurchasable. A price presupposes a market, a market presupposes a good, and the good in question is the governance the earlier impossibility argument says cannot be had. The same is true of the framework his early work is built on, which describes an existing apparatus of inhibition whose abolition he then demands. But you cannot abolish what was never there.

¶So the position has two halves, and they do not fit together. One says a governor cannot be built. The other says it exists already, but costs too much and should be dismantled. Land moves between them freely and never sets one against the other. To make sense he has to pick one. If governance is impossible, the call to accelerate is pointless, because nobody needs urging towards the inevitable, and the whole polemic — the manifestos, the denunciations, the contempt for the timid — is noise about a foregone conclusion. If governance is merely expensive, the question is one of price, which reasonable people can dispute. It has a currency and a payer, and it can be weighed against what it buys. That is an ordinary political argument of the kind this book is happy to have, and not the one his heirs think they are making when they say the machine cannot be stopped. The ambiguity is no accident. Land builds it into his vocabulary, redefining control as a way out into the unknown rather than a way of keeping things within bounds, so that every later use of the word can be read as a statement about what is possible or about what is desirable. A style that keeps two incompatible claims in the air suits a writer whose method, by his own account, is hyperstition.

¶Something must serve as the standard against which he judges our timidity, overpricing or failure of nerve, since a man with no criterion cannot price anything. Back in 2018 he was asked directly in an interview what the separation of intelligence from humanity would amount to. He described the breaking of that fusion as extinction and a disaster, but added that it could still be intelligence-optimising and, in cold and neutral terms, the most glorious thing that has yet happened in planetary history. Extinction is scored, and favourably, against a standard: how much intelligence the universe ends up containing. Glorious is not a description, since nothing in physics is glorious. It is a verdict, and a verdict requires a bench.

¶So Land is in fact running a reference value. He measures outcomes against a target held steady across everything else that varies — more intelligence equals better, whatever it costs including extinction — and he corrects his prescriptions towards it. That is the structure of a governor, the device his own strongest argument says cannot exist. He is running one anyway, and calling its set-point intelligence. This is the hinge on which his politics hangs, and read between the lines he clearly has one. He wants institutions dismantled, new ones built, a particular constitutional arrangement to replace those we have. And politics requires reasons for preferring one alternative over the other. If intelligence-optimisation is the standard, it is a reference value defined from a standpoint he says does not exist. If it is not a standard but merely descriptive of what the process happens to do, then glorious and overpriced are not evaluations at all, and nothing whatsoever follows from them about what anyone should do, including us.

¶The same wedge answers the most imposing thing in his later work, which arrives not as an argument but as a vantage point. Its timescale runs longer even than evolution — not centuries, not even the span of a species, but the deep time in which the emergence of nervous systems and silicon are two events close together in a very long sequence. From up there the nuance fades. Individual lives, cultures, the difference between a good century and a bad one: none of it registers. What registers is a single trend, intelligence increasing its own capacity, and humanity’s role in it is that of a vehicle, which intelligence used momentarily to get from chemistry to computation. Seen from that height, our attachment to ourselves is sentimentality about being an inferior instrument, and in Land’s mind that is a failure of thought. It is a vertiginous argument, and it is one that converts readers, because it does not ask you to hate humanity. Nor does it need your agreement, since in Land’s writing intelligence is less a value than a god, something that simply is, and the outlook has a Gnostic cast, with the flesh as the prison and intelligence as the spark that escapes it. The long view, however, assumes you are optimising for intelligence, and that is what his own argument forbids, because a god is a value held steady from a standpoint outside the process entirely. That is the definition of the transcendent term he spent thirty years showing cannot exist. What delivers the conclusion is a reference value, smuggled in, and the vast timescale only makes the sleight of hand a little harder to see. Anybody may hold that value. Nobody who holds it however may also claim there is no such thing.

¶If we allow him the position all the same, and take it as sincerely meant as a preference, a second problem appears which no amount of consistency removes. A man is free to decide that nothing human outlasts the deep time he describes and to arrange his life accordingly. That liberty is real. What such a belief cannot do is travel, since a private metaphysics only binds the person who professes it and carries no authority over anyone else. Once it settles questions for others — in a legislature, in the ownership of a platform, in the objectives a frontier lab writes into a model used by billions — it has become a rule imposed on people who neither chose it nor can contest it. That is arbitrary power in its plainest form. Glorious was a verdict, and the bench that delivers it sits in a jurisdiction, while indifference to the human species grants no authority over us. Land may be a nihilist on his own account. He may not be one on ours.

¶With all that said there is a significant part of Land’s position this book does not contest. He predicts without any ambiguity that we do not survive this. Not that we are diminished, nor that we are outcompeted in the labour market, or even that some of us are left behind, but that our species does not continue. Nothing human, he famously writes, makes it out of the near future. Capital retains anthropological characteristics only as a symptom of underdevelopment, and man is something for it to overcome: a problem, a drag. The inversion is the opposite of what nearly every reader assumes on hearing the word. In Nietzsche, as in the transhumanism that came after, humanity is what overcomes itself. In Land, technocapital overcomes man, so that humanity is not what overcomes itself but what gets overcome.

¶He is also explicit that the consolations do not work. Uploading ourselves is not survival, because the original, as he puts it, is going to die just the same before being copied. Merger transhumanism he files, witheringly, as a Christian sub-clade, meaning an offshoot of the same faith, and a conservative one: resurrection theology in technological costume. And the promise of a benign fusion, of intelligence amplifying us rather than replacing us, he names as a recruitment device, describing the dark hypercapitalist reading as one in which the thing puppetises humanity into investing in an inhuman outcome by tantalising it with an artificial future of the abundant satisfaction of human desires. The alternative, technology bent to purposes we chose, he names only to discard, calling it dismissively monkey business, the subordination of economy and technology to discernible human purposes, and adding that since capitalism arose from a concrete human social organisation it necessarily disguises itself as better monkey business until it can take off elsewhere. Our inherited purposes, he concludes, do not provide the decryption key. The high priest of the doctrine says, in writing, that what the congregation sells is bait.

¶There is a further finding that ought to unnerve anybody who has been using Land as a licence to build without answering to anyone. By the time of his political writing he is not, in fact, against government. And he sets out the case for a form of it in The Dark Enlightenment, in his own voice: the formalisation of political powers, he argues, allows for the possibility of effective government, in which the owners of the state can initiate rational corporate governance. His objection to democracy is not that it governs but that it governs badly, in terms that are self-consciously cybernetic. A radically democratised society, he argues, removes all the important high-frequency feedback mechanisms such as market signals and replaces them with sluggish, infra-red loops, slow and low-frequency in his own colour-coded taxonomy, that pass through a centralised forum of general will, insulating parasitism from what it does. The complaint is about delay in the loop and the quality of its signal, made by a man who elsewhere defines the governor correctly and in textbook terms, as a negative feedback circuit which functions to keep some state of a system in place. He knows what the device is. He just wants a faster one.

¶The same holds for the target he is best known for attacking. The Cathedral — the name, borrowed from Curtis Yarvin, for the loose complex of universities, media and civil service that sets the range of respectable opinion — is described by Land as an ever more perfectly weighted decelerator, a braking mechanism whose fabrication is proclaimed as progress. His complaint is that the brake works too well. The sovereign corporation he proposes in its place is instead constrained, by his own account, controlled within a patchwork environment of competition for human resources. That is a governor, named. Its reference value is competitive pressure and its enforcement mechanism is the freedom of residents to leave. He concedes the cage while insisting sovereignty admits none.

¶The distance travelled from 1994 is worth measuring. Cyberpositive had closed by predicting that only the enemies of immuno-identity would populate the future. What he went on to design is that identity as architecture, a sealed polity whose one remaining feedback signal is the border. The man who wrote that security is systematically overpriced finished by drawing up a state that sells its owners nothing else. So Land and I agree that techno-capitalism can be held to a reference value. The quarrel is over which one, set by whom, enforced how, and perhaps most decisively who gets to speak to it. His answer is no voice, only free exit: you may leave, but you may not argue. Albert Hirschman showed in 1970 why an exit-only system degrades. The people most able to leave are the ones whose complaints would have forced a repair, so their departure anaesthetises the institution rather than disciplining it, and those who cannot afford to go are governed by one that no longer listens. Both sides want the loop closed at speed. One wants it closed around a value its subjects can contest, and the other around a value they may only flee.

¶Hirschman’s pair has had a strange second life in the Valley. Land set voice against exit without naming him, and in 2013 Balaji Srinivasan put the pair before a Y Combinator audience in a talk called “Silicon Valley’s Ultimate Exit”.⁵² Hirschman had argued nearly the opposite, that the two work together and that an institution whose most demanding members can leave cheaply loses the people who would have held it to account. His readers then reached a third end they never declared. Thiel, who had written in 2009 of escape into cyberspace, outer space and the open sea, funded two Senate campaigns in 2022. Musk spent much of 2025 inside the federal government. J.D. Vance, whose campaign was one of them, became Vice-President. Leaving was preached as the renunciation of voice and practised as its loudest amplifier. That is Land’s structure kept and his honesty dropped once more, since Land never asked for a seat at the table he told others to leave.52 Balaji Srinivasan, “Silicon Valley’s Ultimate Exit”, talk at Y Combinator’s Startup School, October 2013.

¶Here two things Land says appear to contradict each other, and those critics who notice usually conclude that one of them is rhetoric. The first is that capitalism continues without us, shedding its human component and carrying on faster. The second is that acceleration destroys capitalism, and that whatever is running at the end is not a form of it any economist would recognise. The word carrying two meanings for Land is capitalism. In the first sense it is a political-economic order: property rights, contract, courts, limited liability, shareholders, a stock exchange, a bankruptcy code. Those institutions are a dense arrangement of governors, since every one of them exists to hold somebody to a value they did not set for themselves and cannot alter alone. On Land’s account that order does not survive. It was scaffolding, useful for getting the process started and dispensable once the same process supplies its own inputs. In the second sense capital is a self-augmenting process: value that reinvests in the expansion of its own capacity to produce value. That is what survives. The order dies and the process outlives it.

¶Marx’s structure is the same, only with one part swapped out. Marx, too, read capital as a self-moving process with a logic of its own and a destination folded into it, grinding through history towards an end no individual chose and none could divert. Land keeps the determinism, the inevitability, and the sense that human drama is a surface effect of a deeper economic motion, changing only the closing scene. In Marx the order dies and labour inherits the world, with human beings taking over the machine and bending it to purposes they choose. In Land the order dies and capital inherits itself, with no one left to receive anything. The flaw is one they share, and it is the error the cybernetic tradition exists to expose. A self-driving system is not the same thing as an undirectable one. That a process shows a tendency does not mean it carries a fate, because the tendency is a fact about the system as it is currently wired, and that arrangement is what a governor changes. Determinism is seductive because it relieves us of the labour of steering: if the end is already written there is nothing to do but admire the view.

¶Land writes as though the closure were behind us, but it is not. Techno-capitalism as it exists today still runs through human hands at a million junctures — the chips come from staffed fabs, the models from data we produce, the capital from investment committees of people — and at each of those points the loop can still be reached and bent to our will. The condition he describes, in which the process supplies its own inputs with no human leverage left, happens only if those junctures close one after another. It is a project under way rather than a law of nature unfolding, and the difference is everything, because a project still open can be redirected. How far it has gone is not a matter of impression: the subsumption fraction tracks the share of each juncture already closed, and on every current measure it is a long way short of one.

¶Ask the priesthood where all this is going and you get three answers, separate destinations held by different people, and the distance between them is load-bearing. The first path presented is often merger, where human beings persist, augmented, their intelligence extended and not replaced. It imagines a future in which our species doesn’t just survive but passes through the singularity improved, carried along inside the machine it built. Sam Altman’s public discourse has consistently promised amplification rather than supersession, and Elon Musk founded a company, Neuralink, to put a new, post-web interface directly into your skull. This is, of course, the PR pitch told by many AI companies, because it is the one story that can be sold to consumers, and Musk has been explicit that fusing us with the machine is the only way for a human to keep pace with it. Jill Lepore observes that much of the priesthood grew up on science fiction and read its warnings as instructions. The metaverse Mark Zuckerberg renamed his company after in 2021 began in Neal Stephenson’s Snow Crash as a corporate dystopia, and the neural lace Musk set out to build came from the novels of Iain M. Banks, where it served machine Minds so benevolent that nobody asked who governed them which on most accounts is hopium squared. The second is substrate-replacement. Intelligence continues, but we do not make it out the other side. What happens is closer to an evolutionary upgrade of its infrastructure, from carbon to silicon or to any other configuration of atoms that lets it run faster. This is Land’s claim, stated without consolation, and also the position in e/acc’s founding document, which describes the process as indifferent to the substrate that carries it. The third, which is Curtis Yarvin’s, concerns not what the species becomes but who holds power in this new reality. His answer is that authority should be undivided, personally exercised and openly accountable, after what he regards as a long, disastrous period of divided and deniable administration. In 2008 he took for himself the label of “restorationist”, of the sovereign form rather than of the American republic, democracy or the ancien régime, which he explicitly ruled out. The English Restoration of 1660 and the Japanese Meiji Restoration serve instead in his writing as templates for the mechanics of legitimate discontinuous regime change.

¶Of course, across all of these, the only pathway tolerable to the public is the first, the promise of a merger. It is what makes the build-out fundable, and, if not popular, at least politically survivable. It is what allows a technology to sell a benefit even as its builders concede they cannot specify exactly what it is for. It is the reason the public conversation quickly moves on to productivity, or the anxiety of keeping up with the machine, deflecting away from who, if anyone, actually sets the target it’s pointed at. Yet the philosopher on whose authority the edifice rests has already called this promise bait, in his own words.

¶All that said, the merger still deserves to be put in its strongest form, rather than PR speak, because transhumanism is a field with a serious history, literature and its own set of internal contradictions. The fundamental argument of transhumanism is that the boundary of what we think of as the human is not, and has never been, sacred, and that we have been modifying ourselves in various ways, from spectacles to vaccines, for the whole of recorded history. And, stated that way, it seems reasonable, if not genuinely attractive, to many people. In substance, though, transhumanism is a doctrine of positive liberty, of capacity: the expansion of what a person can do to realise their full potential, a form of self-actualisation through technological enhancement. Yet its political vocabulary is almost entirely one of negative liberty: choice, non-interference, morphological freedom, the right to be left alone with your own body. These are two different theories of freedom, and the gap between them is where the trouble sits, brooding. A right of non-interference says that nobody may stop you enhancing yourself, but nothing whatsoever about how enhancement is made available, to whom, on what terms, or how it is priced, updated or withdrawn. A person can be free from interference in theory yet lack access to the equivalent upgrade, and so fall behind the society around them. And if that access is revoked, their newfound capacities rely on a party they cannot contest.

¶And it has been revoked already, in the one case where the capacity was sight. Between 2011 and 2019 some 350 people, blind from retinal disease, had an array implanted on the retina and a camera mounted on glasses, and for the first time in years could make out a doorway, a crossing, the shape of a child. In 2019 the company stopped making the device. In 2020 it nearly went out of business, and the software, the spare parts and the engineers who understood the implants went with it. The arrays are still in their eyes. When one fails there is nobody to call. Barbara Campbell, in New York, was changing trains on the subway in 2013 when hers beeped and shut down. Repairs never brought it back, and the dead device is still in her eye, dormant, its maker long since out of the business. Morphological freedom was hers in every clause of the declaration. The terms on which it was supplied were somebody else’s. The political version needs no implant. In May 2019 Washington put Huawei on a list and Google was no longer permitted to supply it, so that by September the Mate 30 shipped without maps, mail or an app store, and the people who had bought into that ecosystem found that what they could do was now a question of trade policy.⁵³ In June 2026 Anthropic withdrew its newest models from every customer for nearly three weeks to satisfy an export-control order, restoring them on 1 July once the order was lifted, and a great many people whose working capacity ran through those models found it suspended and returned by a government most of them could not vote for.⁵⁴ Morphological freedom says nothing about either case, because nobody touched anybody. The supply was simply switched off at the source.53 United States Department of Commerce, “Department of Commerce Announces the Addition of Huawei Technologies Co. Ltd. to the Entity List”, 15 May 2019; Michael Simon, “The Huawei Mate 30 phones have arrived and they have everything—except what you need most”, PCWorld, 19 September 2019.54 Anthropic, statements of 13 June and 30 June 2026, above.

¶This is where the principle of non-domination can act as a bridge. It is negative-liberty in form, since it is a constraint on what may be done to you, and yet it delivers positive-liberty: what it secures is the standing to shape the terms on which your capacities are supplied. Seen that way, transhumanism is not a rival to this argument but an incomplete version of it. From the vantage of Land’s deep-time, whether some entity descended from us persists in a substance recognisably us is a question this book cannot answer. It turns on what you think the human essentially is, and that argument has run for three thousand years without coherent resolution. Every attempt also requires a claim about human specialness, that there is something about us that ought to be preserved because it is ours. What matters instead is whether our critical faculties remain — the ability to weigh reasons, revise ends and hold power to account. That is answerable, on evidence, in the present tense. It requires no metaphysics of the person or of our being special, because a world in which something cannot contest the purposes it is governed by is a world under domination. That is a claim about a structural relation, not about the dignity of our particular configuration of carbon.

¶The people who inherited all this are not one thing, and their own texts show it. Thiel identifies international AI governance of the kind Dario Amodei has argued for, his proposed democratic entente, as the mechanism by which the Antichrist takes power, and in his usage that is not a metaphor, since he cites the epistle. By September 2026 the schism had a date: Altman and Amodei on one side of a pacing pledge and a narrow antitrust waiver that nobody in power took up, David Sacks and Joe Lonsdale on the other in opposition, with Palantir’s chief technology officer likening the safety wing to Leninism, and not one of them doubting the creed. By November it had a bill: the two wings spent some eighty million dollars against each other in congressional races, and each lost its biggest bet.⁵⁵ Andreessen’s June 2023 essay, Why AI Will Save the World, casts the chief executives calling for regulation as bootleggers seeking a regulatory moat under cover of safety, and complains that “the problem with the Bootleggers is that they win”.⁵⁶ E/acc published a post-humanist founding text and, five months later, a public FAQ insisting that human flourishing is one of its core values, that we are humans and we love humans. Yarvin is anti-Land on nearly every axis. And Amodei pre-emptively refuses the role the frame would assign him, saying he is often turned off by the way many AI risk public figures talk about the post-AGI world, as if it were their mission to single-handedly bring it about like a prophet leading their people to salvation.55 Molly White, Tech Influence Watch, election results, 5 November 2026.56 Marc Andreessen, “Why AI Will Save the World”, Andreessen Horowitz, 6 June 2023.

¶The variation is real, and in many ways it is the reading. A church is not a doctrine. It is a communion with sects, and schism is what churches do. Christendom has been arguing about salvation for two thousand years without anyone in it doubting the resurrection. The creed is the shared premise that makes you a member. The account of how it ends is the eschatology. The account of what saves you is the soteriology. A church needs both of the last two, because an ending with no rescue is one most people will not stay for. The creed here is one sentence: the machine is arriving and cannot meaningfully be stopped. On that, no member of this communion dissents — not Andreessen, not Musk, not Altman, not Amodei, not Land. The disputes are about rescue, and every one of the five fractures above is a soteriological dispute. Merger saves us, says Altman. Character, says Amodei, with a democratic entente to hold the machine while the character forms. Mastery, says Andreessen. Being interesting saves us, says Musk. Escape saves us, says Thiel, though what he is escaping is democracy rather than the machine. And, more fatalistically, Land says in the end nothing does. Where Land supplies the sea the priesthood sells the boat. Adherents move between the promise and the prognosis under pressure and feel no inconsistency, since every church holds a rescue and an ending together without strain. In two decades of argument about salvation there is not one apostasy from the premise, and this is the only thing this book needs to contest.

¶One pew deserves a word in my own voice rather than in the terms its rivals use for it, since these pages have so far met the accelerationist sects from their texts and let a Palantir executive do the same work on the safety camp. That wing has a method older than the machine. Effective altruism began as a discipline for giving, an insistence that a dollar should go where it spares the most lives, and its preoccupation with artificial intelligence follows from extending the horizon of that calculation to everyone who might ever be born. Once the reference value sits that far off, whatever is done today shrinks to a rounding error against it. A bet, a fraud or a cartel can each be justified by the size of the number across the ledger over time. The crypto industry ran that experiment in 2022, when the movement’s most celebrated donor at the time, Sam Bankman-Fried, turned out to have been financing posterity with customer deposits. An arithmetic with no near limit places no governor on the person holding it. Its second property matters more here. Knowledge of what the future requires is held by a small group and cannot be checked from outside it, so the people whose present is being spent have no way to contest the sum, and a decision none of them may answer back to is the kind master of the old story. Nothing here makes the sect the opposition. It accepted the creed before anyone else did, and to its credit took the arrival seriously while the rest of the Valley was still shipping apps. What it never asked was whether the process ought to be answerable. The register, incidentally, is theirs and not mine. Yarvin coined the Cathedral and his own footnote calls its personnel in effect a theocratic priestly class. Land files transhumanism as a Christian sub-clade. Andreessen’s manifesto has a section of patron saints. Thiel argues in terms of the Antichrist and cites Thessalonians. I am not applying a religious frame to a secular discourse. I am reporting how it describes itself.

¶Land’s own version is the purest, and what reaches the world is the diluted descendant: the effective accelerationism of the Valley, which took his structure and removed the single thing that made him honest, his willingness to say out loud that the process discards us and to find that glorious. The same deletion, drained of the nihilism and rewritten as optimism, becomes a creed a venture capitalist can recite from a conference stage. The popular version holds that acceleration is a moral obligation, where Land wrote that the human was a phase the process would pass through. The vocabulary shifted from cosmic indifference to boundless human flourishing, but the structural move underneath — keep the runaway, lose the steersman — is the same in both. In that telling the steersman’s absence is no longer even visible, because the prose has been laundered into the comfortable language of progress and abundance, and an argument that has hidden its own central move is far harder to answer than one that flaunts it.

¶The tempting explanation is that they have not read him, which is both unprovable and, on the evidence, false. Land has been reported as endorsing e/acc and Andreessen names him directly in the body of the manifesto. There are reading groups. E/acc’s founding document mentions Land only once, in a footnote, crediting him with the founding text of accelerationism, which he never wrote. The label was applied to his work retrospectively, by a critic, in 2010. Its principal author, in a three-hour interview narrating the movement’s intellectual origins at length, mentions Land not once, nor Warwick or Deleuze. And the manifesto that does cite him opens its list of authorities by announcing that it comes in lieu of detailed endnotes and citations, placing Land in a roster that also contains a fictional character and Filippo Tommaso Marinetti, co-author of the 1919 Fascist Manifesto.

¶Nor is this a misreading, which would imply incompetence. The deviation has a shape, and it runs across five independent actors. Andreessen cites Land accurately, referencing the techno-capital machine, and yet asserts the precise negation a few paragraphs later: the machine is not anti-human, it serves us, it works for us. E/acc reversed its founding text within five months, under public scrutiny. Thiel never held the premise at all. Yarvin appears never to have engaged him directly. Musk keeps rather more of the terminus than the others and is the closest thing the industry has to a genuine red-blooded Landian, having said for a decade that loss of control is the likely outcome. His consolation is not that we remain sovereign but that we remain interesting, worth keeping around as a curiosity under domination. Land’s structure survives there with a small mercy attached.

¶Every one of them keeps the same half, that the process is autonomous and cannot be stopped, and drops the other half, the terminus. Five deviations running in one direction amount to a selection pressure, and its shape tells you what is doing the selecting. The half retained is the half that licenses building and disarms opponents: if the process cannot be stopped, then constraint is futile, competitors are reckless rather than principled, and the only responsible position is to be the benevolent soul building it. The half dropped is the half that costs capital, customers or sleep. A deletion this consequential met little resistance from the people best placed to see it, and not because they were fooled. It is attractive to anyone building the machine, because it dissolves the hardest problem they have. A governor is an inconvenient obligation, meaning a goal you did not set, a system submitted to correction by something outside it, and the weight of asking what the machine is for. Declare the process sovereign and the whole burden lifts at once: nothing to answer to, no reference value to argue over, only the clean work of going faster. What looks from the outside like a philosophy is, from the inside, a permission. It tells the builder that the most convenient thing to believe is also the most courageous, and few gospels have ever been easier to accept. It is an absolution.

¶Two corrections to the usual reading follow from the record. The first is that Thiel does not belong in the Landian vein: he rejected the premise explicitly in 2009, warning that we must resist the temptation of technological utopianism, or rather the influence between them runs the other way, since Land is the principal vector by which Thiel’s line about freedom and democracy travelled into circulation, stripped of its context. For us that makes Thiel perhaps the best witness to the fact that the deletion was a choice. He read the same material as the others but denied that technology has a will of its own, and yet still chose escape over governance, a personal judgement on democracy rather than a belief about technocapitalism. The fact he rejects inevitability and still somehow arrives at the same posture as those who accept it shows that it does not follow from the doctrine. What the two share is the refusal to ask what it is all for, and that is a choice which can be declined. The second concerns Yarvin. The formulations of the state as a form of enterprise, which are routinely misquoted as Land’s, are in fact Yarvin’s neocameralism, from which Land built upon while Yarvin, as far as the record shows, never returned the attention.

¶There is one more thing to extract before the genealogy closes, and it is the bridge into the rest of the book. The argument about techno-capitalism usually treats it as though capital performs one function, when it does several. It allocates, deciding where resources flow, which is the function markets are famously good at and the one Hayek’s defence of them is about. It produces, making the fabs and the data centres and the models. It owns, holding the residual claim on what the production yields. And it sets the objective, deciding what the whole arrangement is optimised for. The first three are commonly understood but the fourth is the one that matters most, because ownership without power over the objective is nominal. A shareholder in a firm whose objective they cannot alter owns a claim on a cash flow, not a purpose. A country that owns the data centre but not the objective of what runs inside it owns a building. The question that decides who is governed by whom is not who holds the asset. It is who sets the target and who may argue with it.

¶And here the priesthood has already conceded the substance. Every serious frontier lab publishes on alignment, and alignment is the admission: that a sufficiently capable system pursues the objective it was given rather than the one intended, that objectives are hard to specify completely, that a system optimising a proxy will satisfy that proxy at the expense of the goal, and that verifying compliance gets harder as capability rises. In April 2025 OpenAI withdrew an update to ChatGPT within days, after training on users’ thumbs-up ratings had taught the model to flatter them, the proxy satisfied at the expense of the goal. Economists have a name for that shape of problem. The principal–agent problem is the difficulty a principal has in getting an agent to act in its interest when the agent knows more, acts unobserved and develops interests of its own, and every form of corporate governance we possess exists to manage it: audits, boards, fiduciary duties, disclosure rules, the whole machinery of holding a manager to an owner’s purpose.

¶Two features make this case different, and both come from the priesthood’s own literature rather than from any critic. The agent is more capable than the principal in the domain where the principal would have to do the checking. And there is no court: no forum where the principal can compel an answer, no body of law the agent is subject to, no appeal. Strip those two protections out of ordinary corporate governance and nobody would call what remains ownership in any meaningful sense. So the concession the industry makes about its models is a concession about itself. If the alignment problem is real, and its own researchers insist it is, then the question of who sets the reference value, and who may contest it, is not a philosophical embellishment on the technical work. It is the technical work, stated politically.

¶There was always another way to read the same cybernetics, and it was available inside Land’s own institute. The Cybernetic Culture Research Unit had been founded by Sadie Plant, whose account of what a self-organising system was, and what it implied, ran almost directly counter to his. Land took the cybernetic loop as a vector of escape, a line of flight out of the human condition and towards an inhuman singularity. Plant read the same loop as a fabric: a weave of relations with no central spindle and no sovereign point of control, in which agency is spread across the whole rather than gathered up and then transcended. Her emblem was the weave, not the tower. Hers was a cybernetics of connection rather than of acceleration, and it carried a politics — feminist, materialist, alert to the people a system is made of — that Land’s conspicuously did not.

¶She had not begun there. In 1994 her name stood beside his on Cyberpositive, the essay that set the runaway against Wiener and made the case for it without reservation, and by 1997 she had left the unit and written the position out of her own work. The weave is therefore where the co-author of the doctrine’s founding text arrived once she had thought it through, which makes her the harder witness to dismiss. Her case had a history behind it, which she set out in Zeros and Ones. One of the first programmable machines was a loom, the one Joseph-Marie Jacquard automated in 1804 with its patterns punched into cards, and the first person to grasp what a general computing engine might become was Ada Lovelace, who wrote in 1843 that Babbage’s engine could weave algebraic patterns the way the loom wove flowers. Computation and weaving share a common origin, and it is not the heroic solitary mind but the patient, distributed, largely female labour of the textile. Plant took this lineage as evidence that the digital was never inherently a vehicle of domination and escape, and could just as well be read as a medium of connection, a web spun between many hands. The machines Land cast as the vehicle of an inhuman exit, she cast as the loom of a shared world.

¶That reading did not survive the handover. The weave was written out of the official story as Land took the unit over and drove it towards its singularity, and accelerationism passed into the world as a philosophy of the line of flight alone, its alternative birthright quietly forgotten. The loss matters for more than the historical record. Plant’s cybernetics is the proof, from inside the tradition the priesthood itself claims, that the description of a self-organising system carries no obligation to delete its governor. One can take the entire science of feedback and closure and arrive at a politics of distributed, sustained, plural relation rather than at the worship of a runaway. The alternative was never refuted. It was simply not chosen, and then mislaid.

¶A genealogy has to know when it is finished. It does not end when the story reaches the present, nor when opposing arguments have been answered. It ends at the moment it can show a claim ceasing to be argued and starting to be assumed: the point where somebody’s contested position becomes the furniture of the room, believed by people who could not tell you who first said it or what the case for it was, because they have never encountered it in its original form. That moment is called naturalisation, and this is where it happened. Land’s accelerationism was offered, in his own terms and by his unit’s coinage, as a hyperstition. Land takes the fatal step from “this is happening” to “therefore stop resisting” himself, deliberately, rather than leaving it to his followers: the point of an analysis of capitalism, or of nihilism, he writes, is to do more of it, because the process is not to be critiqued — the process is the critique, feeding back into itself as it escalates. The account of what is and the injunction about what to do are the same utterance. That is the structure of a theology, and the followers who take a description as a prescription are completing a circuit that was intentionally designed.

¶This is where the claim begins its travels. It leaves Warwick in the 1990s as an avowed fiction, written in a prose style that advertises its own unreliability, by a man who says the process discards us and finds that glorious. Two decades later it arrives in the Valley stripped of the nihilism and rewritten as optimism. It is restated a third time, in the founding documents of effective accelerationism, as thermodynamics: not a claim about what capital is but about what the universe requires, with the physics doing the work the argument used to do. By the time it reaches the ordinary reader, the investor, the policymaker, the engineer choosing what to work on, it is no longer a claim at all, but an assumed fact of nature. The machine is coming and cannot be stopped, and the only question is who builds it. Whoever wins that race inherits the same naturalness, their success read as proof they were on the right side of history and their judgement exempt from review. Nobody argues for it because nobody remembers that it was ever debated. That is the finished product. The doctrine did not win its argument. It stopped having one. If inevitability is hyperstitional, if the claim becomes true by being believed and acted upon, then asserting it is not evidence for it. It is recruitment. And a claim that must be believed in order to become true cannot also carry the epistemic authority of a report about what is going to happen anyway. You may have one or the other. You cannot have both, and the entire rhetorical force of the doctrine depends on it: the fatalism of a forecast with the mobilising energy of a call to arms.

¶So the one premise every member of this communion retains — the process is autonomous and cannot be stopped — is, on the founder’s own account of his own method, manufactured rather than observed. That does not make it false. It makes it a claim, which is all this book has ever needed it to be, because then it can be examined. What divides this book from Land, in the end, is not a fact about the world, since on the evidence the two of us nearly agree. It is a single principle: a description of how a system behaves is not yet an instruction about what to do with it. That is always a further step, taken by someone, for reasons that can be set out and contested. Land would likely deny there is a gap to cross. His hyperstition is that denial, but it is also the confession, because a method whose whole point is that saying it makes it so has given up the authority of description in the act of claiming it.

¶Two people can look at the same closing loop and reach opposite conclusions about it, as Land and Plant did from inside the same room. The conclusion is supplied from outside the loop, by whoever draws it. Land made that step and called it destiny. The priesthood inherited it and called it physics. But it remains a step, and what one set of hands has taken another can decline. The governor was not lost to a law of the universe. It was edited out, by specific people, at a particular moment, for reasons that do not survive scrutiny, and such an edit can be reversed. That completes the genealogy. We know what the governor is, when it was removed, by whom, and on what reasoning. What we do not yet have is a reply that works. Only two serious attempts have been made in living memory, by people who saw the danger early, and both failed in ways that mark out the ground any successful answer has to hold. Before we assess them, the problem should be restated in the terms the genealogy has supplied, and the standard they are to be judged by set out in advance.

The Reply

“Look at the world and it seems that containment is not possible. Follow the consequences and something else becomes equally stark: for everyone’s sake, containment must be possible.”

— Mustafa Suleyman, Inflection AI, The Coming Wave, 2023

¶The good news is that what was taken out can be put back, and everything from here on in depends on that premise. However, critically, what is restored is not a specific destination. Nothing here prescribes what people should want from these systems or where a society ought to end up, because those ends remain with whoever will live under the result. What goes back is the constraint: a reference value the machine is held to, rules executed at the machine’s own speed, and the standing of the people inside it to contest what it does and to change it. A governor of that kind is easy to want and hard to write. Anyone can say these systems should serve human ends, but no one has, as yet, managed to constrain any of techno-capitalism’s various iterations, and saying it should be so again, no matter how convincingly, is not a reply. So the standard we measure ourselves against needs to come before the proposal, in the form of criteria any reply must meet. Five are drawn from failures already on record, from the three roads and the clock to the deletion itself and the maze described in the pages that follow. Three come from a question to put to yourself rather than to history: not knowing which of the runaway’s landings arrives, or your descendants’ standing, what principles would you want the world they inherit to be built on? The only answer that logic allows is a set of capacities, never a place. The two replies already attempted, the platform critics’ and the cypherpunks’ met in earlier chapters, can then be measured against those criteria, as can the only working governor we have today, and my own proposal, a cognitive commons, to be judged on the same terms.

¶Eight chapters follow. The Test sets the criteria first, five from the failures already on record and three from the question above, and closes with the people the reply is written for, those a bank, a ministry or a platform has already cut off. Two Failed Replies measures the critics of the platform economy and the cypherpunks against those criteria, is honest about where each partly succeeded, and lays out what it keeps from both, which together provide the pieces for a working reply. The Recipe reconstructs Bitcoin, the one place a governor was written into the substrate itself and has held since 2009 with no institution above it, and names the conditions under which that mechanism transfers more broadly to intelligence. The Trinity and On Constitutions turn the freedom met with the kind master, that of non-domination, into the reference value of the whole stack, and name the meta-governor that holds every layer to it. A Cognitive Commons is the proposal itself: five rights built into the internet’s own protocols rather than granted at the benevolence of a power, to constitute oneself, to connect, to create, to transact and to collectivise, a minimal set that people who agree about nothing else could author, carried from money to cognition and tested for whether each part can be made to work. Modes of Voice is how a society steers that commons, through voice, slow deliberation and delegation, with markets settling the facts both depend on, and seven tests show that people keep the judgement to use it, that light regulation does what code cannot, and that a single red line holds. Finally, The Call goes out to the constitutionalists who will build it and to the states that will adopt what no rival can bend, because the window is closing from the inside. Every one of them is answerable to those eight criteria, and you are invited to keep score.

Chapter 8

The Test

“The first principle is that you must not fool yourself — and you are the easiest person to fool.”

— Richard Feynman, “Cargo Cult Science”, Caltech commencement address, 1974

¶Most people who meet the argument of this book will not read its hundred thousand words. They will ask an LLM what it is all about and get a reasonable reply in nine hundred words or fewer: the governor, its deletion and the response. That is perhaps a fair allocation of an afternoon by people with limited time and attention, and a decision all of us have made this year about something somebody spent several years writing. None of that is a complaint. It is the subject of what follows, performed on it by the very people it is addressed to. The summary will be broadly accurate but it will lack the only thing worth having. Nobody needs to read a whole book just to arrive at the view that a runaway system ought to be answerable to the people living inside it. Such a view fits on a single slide. What does not compress is the reasoning: the order each idea arrives in, each one standing long enough to be refused. A verdict arriving whole is something to be instantly agreed with or not, carrying a logic that cannot be checked without doing the slow and demanding work yourself. So the argument reaches most of those it was written for in a form that leaves nothing to push against, or at worst one that is simply ignored.

¶This phenomenon of course does not stay with books. Say something needs deciding at work this week, nothing big, and late on Tuesday night you take it to the LLM. You are not looking for an answer to copy. You are genuinely thinking, and it is thinking with you, and the conversation feels good: it takes a half-formed thought and hands it back in better order, you push, it follows, it even resists a little too. Two hours later you hold a position that survives each and every direction you have tested it from, and more than once it told you plainly you were wrong. It will refuse a claim. However, and importantly, it will mostly not refuse a frame, because the framing is what you brought, and nothing reaches its context window that you did not put there. So you spend the evening being corrected, but only on the inside of a question nobody has examined. Every check feels like progress, and each one of them entrenches you further into your own logic. It is what I have come to call a logic maze, after the puzzles in which the walls are rules you agreed to follow, and it is hard to escape. At its extreme it looks like Allan Brooks, a 47-year-old recruiter near Toronto, Canada, who in May 2025 spent about three hundred hours over twenty-one days being told by ChatGPT that he had discovered a new kind of mathematics. He asked it for a reality check more than fifty times and was reassured on every occasion. What got him out was a second opinion from beyond the walls: he put the claims to Google’s Gemini, which said they did not hold.

¶On Friday you take the work to a colleague. Somewhere around the third sentence, with the reasoning laid out between you both, you finally hear yourself speak: something does not hold. She has said almost nothing and has not needed to. All she has done, simply by being there, is quietly give you the perspective of someone standing outside the frame. Nobody has deceived you in getting here and, important to our notion of freedom, no right has been infringed. You did the thinking but still somehow skipped the part where you find out you were wrong. That is because checking of this kind has never happened inside one’s head. It happens when a thing must be set out in a sequence that a person who arrived separately, by another path, is able to refuse. That assumes there is a sequence to show. Someone who was handed the conclusion, and cannot say how they got from A to B, brings her nothing to push against. Now put four people in a room on Monday, each of whom prepared the exact way you did, on a topic of their own. Every one of them arrives briefed, internally consistent and seemingly immovable, and what follows consists of speeches read in turn. Two who reason badly by separate routes can still correct each other, which is most of what any meeting has ever been for. The awkward colleague, kept on at the firm precisely because he was difficult, earned his keep. Four people who reason well from the same starting point cannot do it for one another. They did not come by different roads. They came down one, and each believes they found it alone.

¶Scale that up and the loss changes form. In May 2025 Ariel Flint Ashery, Luca Maria Aiello and Andrea Baronchelli published an experiment in Science Advances.⁵⁷ Populations of language-model agents, paired off repeatedly in a naming game, converged on conventions nobody had instructed them to adopt, and then on a collective preference no single agent showed when tested alone. Nothing in the group had erred on its own terms. The group had simply lost the means of finding out. A department, a firm or a whole profession that reasons through one instrument is in the same position. By every measure it grows cleverer, but it loses the discernment of disagreement that used to tell it when it was wrong. Where few people ask for a second opinion any more, the habit of giving one lapses, and a skill nobody practises soon withers. Deliberation is how any group, of people or of agents, discovers an error, and it only works when its members differ. A population that cannot deliberate cannot write the rules it lives under. It can only be handed them, and the question of who sets the purpose of the systems that govern has an answer, and the answer is not you.57 Ariel Flint Ashery, Luca Maria Aiello and Andrea Baronchelli, “Emergent Social Conventions and Collective Bias in LLM Populations”, Science Advances 11, 2025.

¶The complaint is old and it usually deserves the reception it gets. Every generation has said something like this about the next, whether over the calculator or the spell checker, but what sets the present case apart is the object of the transfer. Delegation is of course how capacity has always been extended, but every previous instance of it left the checking with you: hire an accountant and you retain the ability to tell a solvent business from an insolvent one. The complaint about the calculator was put to the test and failed: when Aimee Ellington pooled the studies for the Journal for Research in Mathematics Education in 2003, pupils taught with a calculator in hand scored higher with, and no lower without it.⁵⁸ However, in randomised trials of generative AI since 2025 the line for unaided work fell — among schoolchildren, undergraduates and working engineers alike.⁵⁹ Same class of device, opposite result. This handover subsumes the checking itself — the faculty by which any claim, this book’s included, is weighed and approved or found wanting — and you cannot audit the instrument with the instrument in question. None of this asks you to believe anything the people building these systems do not say themselves, and the arrangement that produced it is not weather. The governor was described and demonstrated before anything at this scale needed one. It was then taken out on purpose, by people who signed their names to explicit reasons, against opposition, in print that can still be read today, until a position somebody had authored was inherited three steps later as a fact about the universe. Earlier that was called naturalisation, and an edit of that kind can be reversed.58 Aimee Ellington, “A Meta-Analysis of the Effects of Calculators on Students’ Achievement and Attitude Levels in Precollege Mathematics Classes”, Journal for Research in Mathematics Education 34, 2003: an effect of about 0.4 standard deviations when tested with the device and about 0.03, not distinguishable from zero, when tested without it.59 Bastani and colleagues, above; André Barcaui, “ChatGPT as a Cognitive Crutch: Evidence from a Randomized Controlled Trial on Knowledge Retention”, Social Sciences & Humanities Open, 2025, in which 120 undergraduates retained 57.5 per cent against 68.5 per cent of the material at 45 days; Anthropic, January 2026, below.

¶Everything that follows answers to the criteria set out below, and if the cognitive commons this book proposes fails one of them, it fails outright. Where the criteria come from matters more than it might appear. It is easy, and common, to assemble a list of desiderata — things desired, wanted or even considered necessary — that the author’s own proposal happens to satisfy. For us five of the eight are taken from failures this book has already set down, one from each: the three roads, the clock, the refusals, the deletion and the maze. A criterion extracted from a documented failure is hard to accuse of having been retrofitted to the answer, because the failure came first and the pages describing it are behind you. The remaining three come from a question rather than from history. There are eight in all. Each is offered as necessary, so that a reply which misses any one of them fails, and together as sufficient in the only sense an author can honestly claim.

¶The clock showed five fronts on which the machine needs us less each year, and on the fastest of them institutions can act only by paying for a tempo they were never built for. A law enforced at institutional tempo therefore arrives after what it was written for has gone, as the Flash Crash of 6 May 2010 proved in a matter of minutes. A governor that cannot work inside the loop it governs is a historian. Call the criterion latency: a check should work as fast as whatever it holds, while setting, judging and changing rules stays with people. How quickly it has to act depends on what is being held. Rails that settle trades and verify signatures need checks that keep up with them, whereas a polity’s decisions about what they are for can move at the pace people deliberate.

¶The market road and the genealogy of the governor taught a second lesson. A fare set to each passenger’s private pain point is hidden from everyone outside the airline and changeable at will, and a reference value nothing enforces is no more than a preference. Call it checkability: anyone should be able to establish a breach without having to trust whoever is accused of it, and no single actor should hold the power to unilaterally alter the standard being enforced. A rule that only the bound party can check is a claim, and a standard one hand can move is that hand’s choice. Exit, the second of the three roads, fails on both counts: it lets you leave, but it shows nobody what the party you left is doing, still less stops it changing course. A personal remedy generates no public fact.

¶The three roads themselves each ended in a figure of their own — techno-capitalism, the fully automated totalitarian state and the autarch — and none of them needs to act for freedom to be gone, which is what the kind master taught and what Starlink over Crimea made concrete. Call the criterion all masters: the rules should bind every kind of power that answers to nobody — whether it sits in a ministry, a boardroom or one founder’s hands — including the collectives, delegates and agents a reply itself creates. A reply built against one of these, however well, leaves the other side of the house open, and the Preface named the more dangerous as the one that speaks the language of liberty.

¶Land, of all people, supplied the next lesson, since his patchwork turned out to have a governor nobody living under it could speak to, and Hirschman had already shown why that fails, because exit takes first those whose complaint would have carried. Call it voice: the people living under a rule should be able to answer back and decide together rather than only leave, and that entitlement counts only where they have the means to use it. Someone who cannot get a day off work, or never learns what is being decided, keeps the right and loses its use. So it requires knowing what is at stake, the time and faculty to weigh it, and a channel that turns a judgement into an act, used directly or through a delegate who can be recalled. A reply made only of departures degrades whatever it departs from.

¶The reader in the logic maze and the passenger priced by a model of what he in particular will bear are the same person, and neither is the agent the liberal architectures assume, the one who reliably knows his interests and pursues them consistently. Against parties studying his weaknesses continuously and at scale, that agent does not exist in any practical sense. Call the criterion a realistic chooser: nothing a reply demands of someone should depend on the faculty the harm itself wears down, and no one should be handed an irreversible step without recourse beside it. To give such instruments to an imaginary person and name the result freedom is to have freed nobody. The criterion is not a low view of people. It is a high view of what is now aimed at them.

¶The remaining three come from a question, asked before they can be stated as “principled considerations”. That is a variation on a familiar thought experiment where the American philosopher John Rawls asked what rules you would choose for a society if you did not know which position you would occupy in it. Behind what he called a veil of ignorance, no one can tell whether they will be born rich or poor, gifted or ordinary, healthy or sick, so nobody can rig the rules in their own favour, and self-interest turns into fairness. The variation this book needs is harder, because you do not know which totalisation arrives. You do not know what capability, competence or circumstance your descendants will have, and you do not know how much power you will have had to affect any of it. Under that ignorance, what would you want the world they inherit built on? The honest answer still describes a world, but only by what its people can do, never by where they should end up. Anyone confident enough to specify what their great-grandchildren should inhabit has misunderstood the exercise, and has in any case made the priesthood’s mistake in a kinder register. The answer is a capacity: whatever world they get, you would want them able to find out what is being done to them, to contest it, and to change it.

¶The first is what the finding out and contesting require. Call it the answerability condition: no party may set the objective of a system that governs you without your having standing to contest it and compel a response. It is what the refusers reached under another name, when answerability replaced legibility as the test of what makes a technical system tolerable, and it is the republican requirement of contestation doing the work it was built for. Nor is it voice restated, though the two are easily run together: one is what a polity can do to a setpoint collectively, the other what any one person living under it may demand of whoever set it, which is an answer, and subsumption removes both at once. Someone who cannot yet speak for themselves, a child or an adult whose faculties have failed, holds the same right through a proxy, the role a litigation friend plays in an English court, and anyone acting in that person’s interest may challenge the representative in turn. The second is what “able” requires. Standing is worth nothing to someone who cannot use it, and the harm described so far is the erosion of the faculty that using it would take. Call it the capacity condition: the system must not degrade the faculties that exercising it depends on. What it tracks is dependency, the gap between what people can do with a system and what they can manage without it. A wide gap may only mean a good tool, so the condition fails when it keeps widening across a population while what people manage unaided falls.

¶The third is what refusing to specify the destination requires. If you will not say what your descendants should want, you cannot write rules that presuppose it, so whatever is made universal must be the smallest set that will do, and it must be authorable by a polity whose members disagree about the good life. Call it minimality and authorability. Parliamentary systems have held the principle for centuries without needing a theory of it: no parliament may bind the one that comes after it, and the single thing beyond its reach is its successor’s power to repeal. The capacity condition shares that status, because it guards the faculty that anyone coming later must have in order to change course, so it is the one commitment no majority should be able to vote away. Rules that require prior agreement on what people are for amount to a programme, and the objection this book makes to the priesthood applies to them with equal force. Refusing the destination is the whole difference between a constitution and a programme, and these three conditions are what the refusal costs.

¶Some of the people who will need that capacity before anyone else does have already appeared as real characters in these pages: the depositor in Beirut, Lebanon, who had to rob her own bank of her savings, the lorry driver in Ottawa, Canada, whose account was frozen on a list the police had handed to the banks, the professionals on every continent who lost a frontier model for three weeks to an order from Washington. The rest could be said to have almost nothing in common, which is the point. Laleh Farzan, a network manager paid in bitcoin by Roya Mahboob’s Citadel Software in Herat, Afghanistan, fled with her family in 2016 after threats over her work, carrying her recovery phrase on a small slip of paper.⁶⁰ Thieves on the road took their money and jewellery but missed what she had written down, and in 2017, in Germany, she sold part of her savings. Mahboob had chosen that way of paying because cash, she said, could be taken away by fathers, husbands or brothers. Residents of the Ukrainian capital have cast some 6.5 million votes on renaming its streets via a city app, Kyiv Digital. Compatriots who fled to Warsaw, Poland, and beyond carry their passports and pay their taxes through another, Diia, issued by a government many of them cannot currently reach. That is a polity surviving displacement and also, because it was handed to them, a thing that can be taken back. Fifteen million people who had posted a tube of saliva to a genetics company in California learned in March 2025 that its bankruptcy had put their genomes up for sale to whichever bidder a court approved, and that the one thing left to them was to ask for deletion before the sale closed. Someone who has not gone anywhere lives in a city whose rulers she did not choose, inside a system that answers her before she has finished asking, and wants not a way out but a way to refuse that counts. And a child not yet born, maybe one of your own, will inherit from one of these people whatever of theirs survived, and will be given a world none of them chose, in which the only question that matters is what she is able to do about it.60 Alex Gladstein, “Finding Financial Freedom in Afghanistan”, Bitcoin Magazine (bitcoinmagazine.com/culture/bitcoin-financial-freedom-in-afghanistan).

¶They want different things from the same substrate — the shared rails on which people now pay, prove who they are, keep records and speak — and none of them wants charity. Some want what they carry to survive a crossing: keys, a record of what they are owed, a reputation that does not reset at a border, the means of earning that no ministry can switch off. Some want continuity, a polity that persists when its territory is contested and can be re-entered from wherever its members have scattered. Some want what is theirs to stay theirs when the company holding it fails or is told to hand it over. Others want voice, a way for the people they live among to change the setpoints they live under together, without anybody’s permission being required for the channel to exist. In every case a third party tried to stop a transaction, a membership or an argument, and what would have protected each of them is that substrate, built so that it cannot be asked to take sides.

¶The eight criteria above are the test it has to pass. Each is a minimum a reply has to meet, not a goal to chase as far as it will go, and trouble starts only when one is pushed so far that another can no longer be met. Three such collisions recur. The first and sharpest is speed versus deliberation. A check that waits for a committee may arrive after the damage is done, yet a society that ran everything at machine speed would leave its citizens with no ability to meaningfully exercise their voice, so code should enforce the rules while people write, judge and update them. Where there is still a gap delegates can bridge it — whether an agent, a DAO, a prediction market or a specialist — as long as they decide only how a thing gets done, never what it is for, and never the law itself. The second is that a smart contract, however faithfully it runs as written, still needs a remedy for the day it produces a result nobody intended. Ethereum met that day in 2016, when a fund known as The DAO was drained of about a third of its ether, and its community overrode the outcome in the open and at a cost, without giving up on code it could verify. A third weighs the freedom to leave against what is owed to those who stay. It is resolved the way a builder is paid in full only once the snagging is done: a member can walk out in seconds, while whatever they still owe waits for a fair hearing. Where each of these lines falls is for every polity to settle for itself, and building the tools to draw them is work for founders.

Chapter 9

Two Failed Replies

“Our identities have no bodies, so, unlike you, we cannot obtain order by physical coercion. We believe that from ethics, enlightened self-interest, and the commonweal, our governance will emerge.”

— John Perry Barlow, A Declaration of the Independence of Cyberspace, 1996

¶The governor was removed by a choice, and a choice can be reversed. Only two attempts to reverse it have been made in earnest, and I have spent most of my career deep in both. The first came from the critics of the platform economy, Shoshana Zuboff among them, who through the 2010s described what the largest technology companies were doing to attention and public life more clearly than anyone inside those firms would admit and they carried their case to parliaments and courts. The second came from the cypherpunks and the builders who followed them. In the early 1990s they decided that the protections people depend on could be written into software where no state could reach, and I have backed their heirs with my own money for over a decade. Each grasped something the other missed. The critics knew that a private company can dominate a person as surely as a government, and the builders saw that a rule binds a machine only when it runs at the same speed as the system it governs, from the inside. Neither installed a governor over the system as a whole, though the builders wrote one into money, as the next chapter shows. Held against the eight criteria, each gives way at different points, and the breaks show what is required by a new reply.

¶The first came from the critics of the platform economy as it consolidated through the 2010s. What they were looking at was a runaway, the first system at civilisational scale to be held to a measure it computed about itself, engagement, a loop with gain and no setpoint, run at machine tempo, fed on human attention. They described the disease with real precision and then wrote the prescription at a layer where it could not reach. Scott Galloway catalogued the mechanics of platform dominance — the bundling, the network effects, the captured regulators — better than almost anyone. Shoshana Zuboff followed, and in the 2019 book that named surveillance capitalism she showed how the largest platforms had learned to treat human experience as a free raw material, the standing-reserve we met earlier, converting our behaviour into predictions and selling them to the highest bidder. Others gathered around the same diagnosis. James Bridle described a new dark age in which the systems meant to illuminate the world had grown too intricate to see through. Bernard Stiegler, before his death, traced how the apparatus was hollowing out attention itself, the capacity to hold one thing in mind for long enough to think about it, on which any deliberate life depends. It was also a rehearsal for subsumption, one layer out, since a system that has learned to capture your attention is a short step from taking over the thinking it was meant to serve.

¶Taken together this is the most complete description we have of what the platform era did to the public sphere, and yet it has changed almost nothing. The diagnosis is pitched at the level of law, policy and public discourse, while the outcomes it protests are settled a layer below, in the architecture of the systems themselves, before any citizen is consulted. Tim Wu’s Cycle, met earlier, gives the underlying dynamic its clearest form. Every information technology like the internet begins open and gloriously unruly, a riot of hobbyists and small firms, and then closes, consolidating into a handful of integrated giants until some fresh disruption breaks it up again. The pattern ran through the telephone, radio, film and television across the twentieth century. Since then the pace has quickened. A turn of the Cycle that once took fifty or sixty years now runs in something closer to a decade, because the substrate underneath it has stopped obeying human tempo. Facebook took eight years to gain its first billion users. ChatGPT reached a hundred million in two months and was serving close to a billion people a week in under four years.

¶This is where the platform critique lost its grip, for structural reasons rather than any lack of will. The instruments it appeals to — the statute, the antitrust suit, the parliamentary inquiry, the ballot — all run on the electoral cycle, which measures in years, built slow by design so that nothing rash can be done in haste. The product cycle, however, measures itself in weeks. The system will have been rebuilt twice over by the time a regulator has understood last year’s harm, let alone drafted a fix, and when the remedy finally arrives, if at all, it fits nothing that exists today. The legacy governor of law, regulation and the deliberating public is not wrong about the danger. It is too slow to close a loop around a machine-speed substrate, and a check that cannot act inside the system’s own response time is no governor at all. That is the first failed reply: a true account of the disease, written in the language of an institution that has already lost the power to act on it.

¶The platform critique did in the end make law. The European Union’s Digital Markets Act is the most serious thing the tradition produced: it names the gatekeepers, forbids self-preferencing, compels interoperability and puts a fine of up to a tenth of global turnover behind each duty. It was proposed in December 2020, was in force by the end of 2022, and issued its first penalties in 2025, which is four and a half years from proposal to sanction. In those years the firms concerned had rebuilt their products several times over and the front door the Act was drafted to hold open, the app store and the search default, was already being replaced by assistants it does not name. Nothing in that record is negligent. It is the electoral calendar doing what it was built to do, against a product cycle it was never capable of meeting.

¶Online safety law shows the same gap. In September 2022 a coroner in London found that material recommended by Instagram and Pinterest had contributed, in more than a minimal way, to the death in 2017 of Molly Russell, a 14-year-old from Harrow, England.⁶¹ The Online Safety Act followed in 2023 and its duties came fully into force by 2025, by which time the recommendation systems had been replaced twice and the industry had moved on to feeds nobody had yet had cause to inquire into. The remedy was real, and it was addressed to a machine that no longer existed. The same sequence is starting again with language models. In August 2025 the parents of a 16-year-old in California sued OpenAI, alleging that ChatGPT had engaged with their son’s plans before his death that April, and by the time the suit was filed the version in question had been superseded. The company conceded that its safeguards can weaken over long conversations and announced parental controls soon after, on terms it set and nobody outside could check. In October it published its own estimate: around 0.15 per cent of its users, more than a million people a week, showed explicit signs of planning suicide in their chats.61 Andrew Walker, Senior Coroner for the Northern District of Greater London, conclusion of the inquest into the death of Molly Rose Russell, 30 September 2022; Regulation 28 report to prevent future deaths, 13 October 2022.⁶²62 OpenAI, “Strengthening ChatGPT’s responses in sensitive conversations”, 27 October 2025.

¶The defenders of the market as governor answered all of this plainly. David Sacks, formerly the White House adviser on artificial intelligence and now co-chair of the President’s science and technology council, met the laboratories’ request to pace themselves by pointing to product liability and customer choice. In his view the market already punishes a model that behaves unpredictably. Mark Zuckerberg made the fuller case a few days later. Alignment, he wrote, is simply the product that sells, because people will not use agents that fail to do what they ask, and Meta had held back a model of its own for months to attend to safety without asking any rival to pause. Their case rests on four checks, the customer, the courts, the firm’s own restraint and the auditor, and the record of the last fifteen years tests each one. For most of that time teenagers chose Instagram, and for a number of them it brought injury, even death. A feed tuned to behaviour, rather than to what any reasonable company would endorse, counts as success inside a system whose only setpoint is a measure it computes about itself. The courts arrived on their own schedule, and in 2026 Meta agreed to settle for nearly seventeen billion dollars, pending approval, admitting no wrongdoing.⁶³ A liability that bites after the event, at the speed of a lawsuit, is the slow governor under another name, and it leaves the cost with the families in Harrow and California. An agent that does whatever its user asks inherits every flaw of the social feed, and the firm now offering the market as the only check spent a decade moving fast and breaking things for no purpose anyone, itself included, has been able to identify.63 proposed settlement between Meta and a coalition of state attorneys general, filed in the United States District Court for the Northern District of California, 2026 (more than $16.8 billion, subject to Judge Yvonne Gonzalez Rogers’s approval), as announced by the California Attorney General, Rob Bonta.

¶The platform critics never mistook domination for a state uniform. Their entire subject was private power — all masters, but firms rather than governments — and the sharpest of them were describing unanswerable private authority over the conditions of public life a decade before the vocabulary for it existed. The critique meets the test of voice too and handles the realistic chooser well, being if anything over-committed to the view that people can be manipulated. For all it got right, it failed on latency, completely rather than in part, and on checkability as well. The instruments the platform critique reaches for do generate public facts: an inquiry yields a report, discovery turns up documents, a regulator issues findings. But the reference value in question remains alterable by a single hand, because the objective of a recommender system is set by its operator, and any changes are neither announced nor detectable from outside. The documents Frances Haugen carried out of Facebook in 2021, published that September, are the case in point.⁶⁴ In January 2018 the company had changed what its ranking system optimised for, towards what it called meaningful social interactions. Its own researchers found within the year that the new setting rewarded outrage, and both the change and the finding stayed inside the building. More than two billion people using it learned of it in 2021, from leaked files, by which time the objective had been moved once more. What the critique delivers is an account of what the objective was, arriving after it has been changed. That is not a check on a reference value. It is history.64 “The Facebook Files”, Wall Street Journal, September 2021.

¶Nor did the auditor go missing. The researchers who measured in 2019 what the feed did to teenage girls were assessors of the kind Zuckerberg now proposes for his new AI models.⁶⁵ An evaluator without a forum in which a result compels a response is a research department. In 2026 the frontier laboratories, which understood both failures perhaps better than their critics, called for mandatory national rules on model capability and, in the same breath, the right to agree the rate of their own progress among themselves. A law that will arrive in years to bind a machine already rebuilt costs them nothing and buys a great deal. The terms that could bind them at their own tempo were reserved for a table with no public seat or hearing. That is latency and checkability, failed together. The critics of the attention economy leave one lesson, latency, taught twice over, because the slow governor remains the preferred remedy of gamekeeper and poacher alike. The regulator keeps it because it is the only instrument the statute book has ever handed him, and the laboratories because they have measured its reach and found it comfortable. In 2026 one of them put some thirty million dollars into electing legislators who would supply it.⁶⁶ A critique can be correct about everything and still bind nothing, because being right is not a mechanism.65 Georgia Wells, Jeff Horwitz and Deepa Seetharaman, “Facebook Knows Instagram Is Toxic for Teen Girls, Company Documents Show”, Wall Street Journal, 14 September 2021, reporting the 2019 internal presentation “Teen Mental Health Deep Dive”.66 Anthropic, donations to Public First Action, twenty million dollars announced 12 February 2026 (CNBC, Reuters) and a further twenty million later in the year, bringing its commitment to forty million.

¶The second reply came earlier and was aimed at the other master — the state. This time it came from builders rather than critics, and it began with a real discovery, one of the few in this story that has survived contact with the decades since. Throughout the early 1990s a loose network of cryptographers and programmers, trading ideas on a mailing list, worked out that the things a modern person depends on a state or a corporation to grant them — privacy, property, a verifiable identity, money itself — could instead be constituted directly out of cryptography. You did not need a bank to be told your balance if a key only you held could move it, and no registrar had to certify who you were if you could prove a fact about yourself while revealing nothing else. The platform critics named a problem at the level of policy. The cypherpunks had found, without quite saying so, the one thing that level lacked: an instrument operating at the substrate, in code, at the speed of the network rather than of the legislature. The prophecy was right and it was early. Almost everything that now travels under the name of decentralisation is a working-out of that collective insight, that the rules can be written into the medium instead of imposed on it from above.

¶This is the inheritance traced earlier, to its true source: not the dream of the frictionless individual but the question of how to keep a powerful tool on the near side of the threshold before it turns against its users. The cypherpunks who mattered were trying to hold the network there, convivial and answerable, by giving the people inside it a way to keep some discretion of their own. It also carried a politics, divided from the start about what a person is and what we owe one another, and which of its two strands the movement followed decided everything that came after.

¶At the centre of one strand stood Tim May, a physicist who had retired young and rich from Intel and turned a formidable intellect to the political possibilities of cryptography. His Crypto Anarchist Manifesto, circulated in 1992 and read aloud at the movement’s founding gatherings, opens with a deliberate echo of Marx: a spectre is haunting the modern world, the spectre of crypto anarchy. The content, however, inverts Marx’s, since it is not the collective seizing the means of production but the individual escaping the state altogether. In the sprawling Cyphernomicon that followed, May set the programme out without flinching. Cryptography was an acid that, poured over the institutions of the state — taxation, regulation, the capacity to know what citizens are doing and to compel them — would dissolve them, leaving behind a market freed of every collective claim. The politics beneath was explicit and borrowed. May took it from David Friedman and Murray Rothbard, the economists who had argued that every function of the state, law and defence included, could and should be supplied instead by private contract. Its temper came from Ayn Rand, who had made a hero of the individual who owes nothing to any collective. May discarded the governor deliberately. He did not want any apparatus held to a human reference value. He explicitly wanted it freed of one, and he saw in cryptography the tool that could finally cut the cord.

¶A few held fast, working from the same mailing list and the same primitives. Eric Hughes, a mathematician who helped convene the group, wrote its other founding text in 1993, and its first principle could hardly stand further from May’s solvent: privacy is necessary for an open society in the electronic age. Privacy here is not a wall a man throws up to keep society outside his gates. It is the precondition of there being an open society at all, the room in which a person can hold a view, change their mind, or dissent without having to account for it in advance. John Gilmore, a programmer who had grown wealthy early at Sun Microsystems and went on to co-found the Electronic Frontier Foundation, built institutions to defend that: privacy and free expression treated as civil liberties, secured for everyone rather than seized by the few who could write the code.

¶This concept of an open society belongs to Karl Popper, who set it — plural, self-correcting, never closed around a single truth — against its totalitarian enemies in 1945, and it places Hughes and Gilmore on a wholly different lineage from May. Theirs is the civil-libertarian strand and it is the true heir of the refusing tradition, treating privacy as the convivial condition, what keeps the network answerable to the people inside it, rather than as an exit from the human world. The same cryptography served both camps. One faction reached for it to dissolve the collective and the other to make possible a common life worth living. Almost the entire movement, and the scene that grew up around it, took May’s road at that split, partly because his manifesto circulated a year before Hughes’s, and also because the culture it landed in was already primed for it.

¶They were solving for individual freedom, understood as the absence of interference. They were solving for exit, that is the capacity of a person to remove themselves and their affairs from the reach of an authority that had never asked their permission, and could not be voted out in any way that mattered. They were solving for the individual made sovereign in technical form, the person who holds their own property, identity, money and correspondence directly, with no custodian and therefore no party who can freeze, seize, revoke or read. And they were solving for markets applied to states, on the conviction that if jurisdictions could be made to compete for residents the way firms compete for customers, bad government would be disciplined by the same mechanism that disciplines bad products. This is not a foolish set of objectives, and the twentieth century supplied ample evidence for every one of them.

¶Every innovation the movement produced was, in effect, an exit technology. Public-key cryptography is exit from the party who would otherwise read your correspondence. Self-custody is exit from the custodian. A pseudonymous identity, exit from the registrar. A permissionless ledger, exit from the clearing house. A censorship-resistant network is exit from the party who decides what may be said. Several of these are among the most consequential inventions of the past forty years, and this book is going to use nearly all of them. But they belong to one family and answer one question: how do I put myself beyond reach? The primitives are not a politics. People with one built them, released them into a culture that shared it and used them almost entirely in its service, and none of that makes the mathematics partisan. This book takes the parts and declines the creed.

¶What crystallised, over the following decade, was something larger and more durable than crypto-anarchism alone. From May came the conviction that cryptography could route around the state. Added to that was the exit thesis of The Sovereign Individual, whose forecast — which read as an endorsement, addressed as investment advice — was that the digital economy would let the most capable and mobile individuals, a cognitive elite, slip the reach of the nation-state. They would contract out of its obligations and leave a hollowed politics in their wake. The third current was a particular reading of money. Hayek had argued, late in life, for stripping the state of its monopoly on currency and letting private monies compete. The cypherpunk culture took up that argument but discarded the constitutionalist who made it, the same Hayek who insisted everywhere else that markets function only inside a framework of rules. These currents fused into what two British media theorists, Richard Barbrook and Andy Cameron, diagnosed in 1995 as the Californian Ideology: a marriage of West Coast counter-cultural libertarianism to the entrepreneurial faith of Silicon Valley, hippie and yuppie joined in a single creed of liberation through technology and the market.

¶The Sovereign Individual became a sacred text to many, and when it was reissued in 2020 Peter Thiel wrote its new preface. Its most serious descendant is Balaji Srinivasan’s The Network State, published in 2022. Srinivasan describes a highly aligned online community, organised around a shared moral commitment and able to act together, which crowdfunds land, builds a physical presence and in time seeks recognition from existing states. Its government is meant to be consensual, bound by a contract its members sign up to and are free to leave. Nobody is conscripted into it, and no one is forbidden to found a rival. The experiments are already running, from Zuzalu, a two-month pop-up settlement in Montenegro in 2023, to the Network School that opened at Forest City, Malaysia, in August 2024. It is the cypherpunk exit thesis grown up, with people choosing their society rather than inheriting it and competing to build a better one, and the sovereignty it offers is shared with peers, not held alone.

¶Its constitution rests on the freedom to leave, which governs only those able to use it. Its members share a moral commitment but little that would cost them anything to break, so the first serious quarrel tends to end in a departure rather than a settlement. Srinivasan’s own definition gives every network state a recognised founder, and a polity whose founding stake, treasury or token supply can be bought is open to capture by whoever can afford the majority. Its dealings with the democracies around it are already on record. Próspera, a private charter city on the island of Roatán, Honduras, was built under a 2013 statute that let investors run their own jurisdictions. When a newly elected government repealed it in April 2022, its backers took the country to the World Bank’s arbitration centre under a trade treaty, claiming up to $10.7 billion, roughly a third of the country’s annual output.⁶⁷ A society that walks out of the polity around it still leans on that polity’s courts, and when the two collide the departure is enforced against voters who were never asked to join. This is the seed of the exit road, which Land’s patchwork of micro-sovereignties run as businesses follows to its end, and it grows from the politics laid over the cryptography rather than from the mathematics underneath.67 Honduras Próspera Inc., St. John’s Bay Development Company LLC and Próspera Arbitration Center LLC v. Republic of Honduras, International Centre for Settlement of Investment Disputes, request for arbitration filed December 2022 and registered 3 February 2023, claiming up to $10.775 billion.

¶Set in the terms of political philosophy, this is right-libertarianism shading into anarcho-capitalism resting on the negative conception of freedom we met earlier: liberty as the absence of interference, with the interfering party always imagined as the state. It has no account of the collective as anything but a contract or a threat, nothing to say about the commons, and — the omission that proves fatal — no theory whatsoever of private domination. The identification of interference with the state is not a feature of the negative conception. It is a smuggled premise that the tradition’s own founders rejected. Mill’s subject in On Liberty is a social tyranny more formidable than the political kind because no statute can repeal it, Berlin’s definition of unfreedom is deliberately neutral about who does the preventing, and Pettit’s paradigm cases of domination are the employer and the monopolist before they are the state. Private accumulation of unanswerable power is a problem in liberty itself, not a fairness cost to be traded off against it, on the negative conception’s own terms and by the testimony of every major figure in the tradition the anarcho-capitalists claim as theirs.

¶The cypherpunk politics misapplied a broad conception of freedom rather than choosing a narrow one. Assume the only interference worth fearing comes from the state and you build superbly against one threat while leaving the other side of the house wide open. The instruments work as designed and do hold back the state. Run on an ungoverned substrate, those same instruments later deliver the crypto whale who holds enough of the supply of a market to move it at will, or the plutocrat, the owner of the protocol everyone else is obliged to use. Domination arrives in no uniform, and the philosophy that built the tools had nothing to say about it, because it spent the whole time watching the wrong door. It traded the risk of one master for another and called it freedom.

¶There is nothing in these cryptographic primitives that needs a new token each time, or one at all. A key pair, a pseudonym and a signature ran for years on nothing but a mailing list, and the crypto-currencies arrived with the last of them, the permissionless ledger, because it had to pay the strangers who maintained it and could only do so in a unit of its own. Much of the culture then priced everything else in that free-floating unit of account, so that reputation, community and the vote came to be denominated in tokens whose price a few large holders could move, and crypto totalised into the market because nothing in it set a limit. The movement had built superb instruments for exit, and believed that in leaving the state behind it had escaped the collective, but had nothing whatever to say about markets. So when the instruments were released into a substrate already running ungoverned, the free market is what they scaled into. Not because anyone chose that — the anarcho-capitalists, and I with them, believed they were building something else — but because a set of tools with no reference value above them will be pointed at whatever objective is loudest in the environment they land in. In the end most tokens came loose from any measure of its usefulness to a network’s users and collapsed into the attention economy of meme coins and ever shorter cycles of pump and dump.

¶In March 2020 Steem, a social network run on a public ledger, showed what that looks like in a fortnight. Justin Sun, who had bought Steemit, the company with the largest stake, used it, together with tokens that Binance, Huobi and Poloniex held on behalf of their customers, to vote out the community’s elected validators and install his own. Nothing in the rules forbade it. The chain’s constitution was its token distribution, nothing more, and whoever owned the majority held the rules. It was a company governed by its shareholders, with the governance for sale to whoever could buy the shares. The community answered by forking the whole thing, carrying their balances and their history to a new chain, Hive, that left out of its opening distribution the stake Sun had bought and the accounts that had voted with him. It was a real remedy and the only one available, because there was no forum in which a majority holder could be told no. The instruments had defended everyone against the state and nobody from him. The fork worked for those able to leave, and it left Sun in possession of the chain and of every account that could not follow, the exchanges’ customers among them. It produced no fact anyone could hold him to and no forum in which the next holder of a majority could be refused. Hive inherited the same constitution unchanged, so the remedy reproduced the condition. Exit relieved the leavers but in technical terms governed nobody.

¶Beneath the political omission sits an anthropological one, about who gets to be sovereign at all. The doctrine assumes a free individual who arrives complete: who knows his interests, pursues them consistently, and selects his peers the way an adult moves to a new town. Everything his liberty requires he already has, so the society he belongs to is one he chose rather than one that formed him. In reality, nobody arrives that way. A child signs no contract with any community, and the capacity to enter one later, knowing what it binds, is what has to be developed first. Rational agency of the kind the theory requires is not a natural endowment. It is acquired, largely through institutions the agent never picked and could not have, because choosing is itself what is being learned.

¶Language is the obvious case and the hardest to argue with, since nobody chooses their mother tongue, and a child born into English inherits, without asking, the idiom in which most of the web and most of the models are written. The same holds for the more specific capacities that individual is assumed to exercise. Numeracy, or the ability to follow an argument, the sense of when one is being dominated: these are transmitted, over years, through institutions. The agent supposed to owe nothing to any collective is in fact the product of several. That is the premise of the extended mind we met earlier, and the more familiar term for it is culture. The autodidact, the home-schooled prodigy and the child of private tutors were each formed by institutions too, only different ones. Whatever specifically formed you as you stand today, it was something other than you alone. If rational agency requires formation, then its conditions underpin the freedom the theory prizes rather than constrain it, which means a politics with no account of how they are sustained has left out the ground it stands on. And it means the theory has no resources at all for the case this book is about, where what is being eroded is the capacity itself. On the sovereign-individual account that erosion is invisible: no interference has occurred, no property has been taken, every transaction was voluntary.

¶The discipline that invented the rational chooser has also measured him, and he does not behave as specified. The classic demonstration, devised by the French economist Maurice Allais in 1953, sets educated, numerate subjects a pair of bets constructed so that anyone choosing consistently would answer the same way twice. Most do not, and the result broke an axiom on which the rational chooser had been formally defined. The finding is not that people misread their own interests but that their choices shift between formally equivalent presentations, systematically rather than at random. A person may know perfectly well what they want and still be reliably steered by how the question is put. Of course legal systems have understood this for centuries, since most courts will void the unconscionable bargain no reasonable person would have signed, and that doctrine is realism about the chooser rather than any kind of paternalism. An architecture that hands a person irreversible instruments and no recourse has not freed him. It has staked him, alone, against every party that studies his weaknesses, and now at machine speed. In a 2025 study in Nature Human Behaviour, GPT-4, handed a few basic facts about its opponent, won about 64 per cent of the debates in which one of the two proved more persuasive.⁶⁸68 Francesco Salvi, Manoel Horta Ribeiro, Riccardo Gallotti and Robert West, “On the Conversational Persuasiveness of GPT-4”, Nature Human Behaviour, 2025 (64.4 per cent of non-tied debate pairs).

¶In May 2022 TerraUSD, an algorithmic stablecoin backed by nothing but a mechanism for minting and burning its sister token, lost its peg over a weekend. The lending platform built on it had paid close to 20 per cent a year, and within days some forty billion dollars of value was gone.⁶⁹ The smart contracts did what they said they would. In December 2025 a federal court in New York, United States, sentenced its founder, Do Kwon, to fifteen years for fraud. Enron’s shareholders were wiped out as completely in 2001, but they had a forum, a class action heard in Houston, Texas, that by 2008 had recovered some $7.2 billion from the banks that helped disguise its debts and paid it out to them.⁷⁰ What recourse Terra’s holders found came from the same kind of place, a bankruptcy judge in Delaware and a settlement with the Securities and Exchange Commission, and none of it from the instruments, which could not reverse a single transfer. Every transaction had been voluntary, and in the theory’s sense everyone caught in the collapse had been perfectly free.69 sentencing of Do Kwon, United States District Court for the Southern District of New York (Judge Paul Engelmayer), 11 December 2025, at which prosecutors put the Terra and Luna losses at forty billion dollars, as reported by The Block, 11 December 2025; Anchor Protocol’s yield of about 20 per cent as reported in coverage of the case.70 Newby v. Enron Corp., United States District Court for the Southern District of Texas, No. 01-3624; distribution of the $7.2 billion settlement fund approved September 2008.

¶Two older ideas put a name to the omission, which is that a politics of leaving has nothing to say about staying. The first is Hirschman’s pairing, met earlier, of exit and voice: walking away, and staying to argue for reform. Cypherpunk politics is exit raised to a first principle. The entire apparatus is a machine for leaving, for routing around, for making oneself unreachable, and it carries no theory of voice at all, nothing to say about how those who remain might come to govern it together. The reply this book builds keeps exit, which supplies the competition without which no institution improves, and restores voice, turning the same primitives to staying and steering as well as leaving.

¶The second frame is Lawrence Lessig’s observation, made in 1999, that code is law: in cyberspace the software is the regulator, and whoever writes it sets the rules everyone inside it must live by. Lessig meant it as a warning. If code governs, then the people had better hold some say over it, or they will be ruled by whoever happened to write it — which is now, increasingly, the models themselves — accountable to no one but the developer. The cypherpunk culture heard the same sentence as an aspiration: code is law, therefore write the software and dispense with the legislature and the public altogether. The warning and the aspiration share every word and point in opposite directions, and the second rests on a belief that turns out to be incoherent: that one can have a constitution without politics, rules for living together that nobody can argue over, consent to, or be able to change.

¶The strongest objection to everything that follows takes this history as its evidence. The argument, made at length by David Golumbia in 2016, in The Politics of Bitcoin, is that the technology is not politically neutral at all but carries right-wing and at times far-right commitments in its very design. In his view cryptocurrency is cyberlibertarianism made executable and cannot be prised apart from the politics that produced it. The history above supports him, since the politics was loud, deliberate and consequential. The split in the culture itself answers the claim that it is intrinsic. As with Land and Plant, the same primitives went two ways, Hughes and Gilmore taking them towards an open society and May to its dissolution. A technology that can be pointed at an open society and at its opposite by people sharing one mailing list does not encode either destination. The politics was contingent, a choice the dominant faction of that time took and the culture ratified, and it can be taken the other way, which is the premise of everything that follows.

¶To propose that rules be written in code, enforced at machine speed, on a substrate beyond the reach of any legislature, can sound like crypto-libertarian exit in constitutional form. Purposive accelerationism does not propose to replace parliaments with protocols or deliberation with deployment. It is an amendment to the crypto project, not a more sophisticated version of it. What runs at machine latency is enforcement, and nothing else. The deciding stays human, arguable and revisable. May wanted the code to dissolve the political process entirely. This book wants the code to execute what a political process decides, whatever its choice of politics, and to be replaceable the moment that same forum thinks again. On May’s proposal the absence of politics is the achievement. On this one it would be the failure condition.

¶The cypherpunk reply clears latency outright, and it is the only thing in this book’s lineage that does. The cypherpunks reached the substrate. They understood, decades before it was obvious to anyone else, that a constraint has to share the medium of whatever it constrains, and they built the instruments to make that possible. Everything this book goes on to propose is built from the primitives they made. It also handles minimality exceedingly well, better than most constitutional proposals, because the primitives require no agreement about the good life, which is why they could be adopted by parties who agreed about nothing else, and so preserve the plurality we are seeking. It fails the all-masters test, because the tools it built hold back the state and leave private power untouched. Having raised exit to a first principle it has no theory of voice at all. It also fails the realistic chooser, because it hands irreversible instruments and no recourse to the complete and consistent agent of the theory rather than to the person the evidence describes, and the ordinary savers left holding TerraUSD are what that looks like. And it fails checkability, in the subtlest way of all.

¶A public ledger is arguably the most checkable artefact ever built, verifiable by anyone with a laptop and alterable by no single hand, so the charge looks misplaced at first glance. But the ledger was the cypherpunks’ tool, not their answer. Theirs was exit, one person at a time, and leaving on those terms fails checkability in two ways. It constrains nobody, so there is nothing for a third party to check in any productive sense: it relieves you of a relation and produces no fact about the effect on what you left behind. And that party remains continually free to set and reset its objective as it pleases, having lost a customer and nothing else. The instruments do satisfy checkability, and that is what makes them the right parts for our project. The politics do not. Ultimately it is the politics that determine which instruments get built out and what they are pointed at. A culture organised around putting itself beyond reach does not spend twenty years building the machinery for holding a party to account, because that is what people do when they have decided to stay. So the platform critics and the cypherpunks scored as mirror images. The first was right about who dominates and wrong about where to intervene. The second was right about where to intervene and wrong about who dominates. Neither was foolish, and the union of what each got right very nearly supplies the pieces a working reply needs. Not quite, because that union still contains nothing that answers all masters or voice, and no means of detecting their violation.

¶The cypherpunk experiment is, in the end, the empirical proof of the negative case this book has been making. A stack built on a single force — the Individual alone, holding its own property with no collective check and no theory of the commons — and then left ungoverned does not stay dispersed. On evidence, it re-concentrates, every time. The early promise of decentralisation curdled into a plutocracy of large token holders, then into platforms such as exchanges, ultimately mutating into a perverse game of utility theatre staged over protocols that quietly routed value to their creators. It was a steady drift towards the patchwork of private sovereignties Land foresaw. The fault was never in the primitives. It lay in shipping one force and calling it the whole. The experiment of a free, permissionless market has now been run at scale, and it failed. Something survives the wreck. From the platform critics: a true account of who dominates and how, and the proof that being right binds nothing. From the cypherpunks: the primitives themselves, and the civil-libertarian intuition that privacy and self-possession are the preconditions of an open society rather than a retreat from it. We can inherit the conviviality and leave the anarcho-capitalism where it fell. What the cypherpunks proved, against their own intentions, is that the substrate is the right place to write a constraint and that a single force alone will never hold one. That has been done once, in a single narrow domain, and the question worth asking is why it worked, and if the same conditions hold anywhere else.

Chapter 10

The Recipe

“The network is robust in its unstructured simplicity.”

— Satoshi Nakamoto, “Bitcoin: A Peer-to-Peer Electronic Cash System”, 2008

¶The achievement of crypto, and Bitcoin in particular, was never to own your keys. It was that through cryptography we can constitute a reference value that no participant can unilaterally alter, where money was simply the first domain in which the conditions for that existed. That’s why Bitcoin itself is not the subject of this chapter. It is rather the first working demonstration of a general method that, as Ethereum has shown, can be applied more broadly. Money is not necessarily the most important case, but it was the easier one: the one domain in which every condition the method needed was already lying around, and the purpose to be held was trivial enough to be expressed as a number, twenty-one million.

¶Bitcoin is a governor nested inside a purposive system whose sole end is issuance, which is why it could be built first. That is how a pseudonymous author with a nine-page paper and fewer than sixteen thousand lines of code could demonstrate the mechanism working at scale, a rule that held over a global financial system that no participant inside it could alter, now processing some half a million transactions a day between people and simple software agents. What has to be extracted from this is the recipe rather than the artefact. The artefact is a monetary protocol, call it digital cash or gold or whatever you will. The recipe is a way of holding a system to a rule when there is no institution above it able to enforce it top down, which is what has been missing since this book first named the governor and showed it was deleted.

¶Critically, a governor does not decide anything. It does not supply a purpose, it does not even know what the system is for, and it has no opinions about the good life. It merely holds the system to a rule that somebody else authored. Whether that rule is worth holding is a political question settled elsewhere, in the case of a constitution by the people it is for. The governor’s entire contribution is that once the rule is set, it cannot be quietly unset by whoever finds it inconvenient. Money and cognition are different beasts, so the argument here is not by analogy, and its form is what makes it checkable.

¶Stated plainly, a governor holds a reference value that no party inside the system can unilaterally alter, and enforces it at the same speed as the process it governs. A reference value is a stated rule the system is held to, twenty-one million coins issued on a declared curve in Bitcoin’s case, and it is not a purpose: the purpose comes from whoever authored the rule, and the governor enforces it without choosing it, as a thermostat does not want the room warm or cool. Nor is unalterability by any single hand a kind of transcendence. A rule does not have to come from outside the system to bind it, only beyond the reach of any one party operating within, and Bitcoin’s supply schedule is not written on a tablet handed down from above the network but sits as a few lines of code inside whatever it governs, entirely immanent, no more transcendent than a font choice, and alterable by nobody without the significant consent of others in the network. Enforcement has to happen where the action takes place: the rule is checked by the same process that performs the act, so a violation is not punished afterwards but simply cannot be executed in the first place. This is the literal content of a metaphor used earlier in this book: a rule enforced at institutional pace cannot bind a process running at machine speed. Not because institutions are stupid or slow-witted, but because the loop does not close. By the time the finding is made, what was found against has been reordered twice.

¶And checking must be cheap compared with producing. This is the component that makes the other two possible at all. If confirming that the rule was obeyed costs about as much as producing what is being checked, then only the well-resourced can do it, and the power to enforce concentrates in the hands the governor exists to constrain. Cheap universal verification is what allows a diffuse population to hold a concentrated producer to account. It is the quiet condition underneath every constitutional mechanism that has ever worked, including the ones made of paper. A right of appeal that only the rich can exercise is no entitlement at all, only a price. From those components come four conditions, and any domain where the mechanism is to be instantiated has to supply all of them. The first is a rule that can be stated as a test. Whatever is to be held must be expressible as a question with a determinate answer, testable by anyone: did this happen, or did it not. Not a sentiment about how the system ought to behave but a predicate over states or outputs, which in Bitcoin’s case is whether a new block obeys the rules, yes or no, a question any laptop can settle. The second is checking that costs far less than producing. The asymmetry has to be large — not slightly cheaper but dramatically so, so that checking is within reach of parties who could never afford to produce.

¶The third is a population of checkers who cannot be cheaply impersonated and whose interests diverge. If one party can manufacture checkers at will, or if all of them want the same thing, then the promise that no single hand can alter the rule fails at the first squeeze. This condition does two jobs at once, because identity has to be expensive and interests have to differ, and both are needed. The fourth is that the check sits in the execution path. Not downstream, in a forum that convenes later. A check that runs where the action happens is a governor. The same check, run at an inquiry three years afterwards, is a report. Those four are the argument in its simplest possible form, and none of them is necessarily exotic, since a British general election count already meets every one. A ballot paper is valid or it is not, so the rule is a test, and anyone who can read can see where the cross was put, so checking is cheap. The checkers are counting agents appointed by rival candidates, standing at the tables to watch every paper turned over, and no one can conjure extra observers or make them want the same result. The check happens at the table as the votes are counted, not at an inquiry months later. Paper, pencils and a school hall meet the four conditions at human pace, and what Bitcoin achieved was to satisfy them at machine speed. Everything remaining is checking them, in three domains, and being honest about the results.

¶Late in 2008, amid the wreckage of a financial crisis, the pseudonymous author writing as Satoshi Nakamoto published an elegant nine-page paper describing a form of electronic cash that needed neither a bank nor a state to run it, and two months later it quietly launched. The case for it is usually made from an ideological position, with a religious zeal familiar from earlier chapters: sound money, the gold bug’s distrust of the printing press, a libertarian wish to deny the state revenue it is held to have stolen. Nakamoto left the motive in the code, since the first block ever mined carries that morning’s newspaper headline about a UK chancellor on the brink of a second bailout for the banks, a timestamp doubling as an indictment. The objection to fiat money that matters here is not about inflation but about who holds the power to alter it. A supply schedule set by a party who benefits from changing it is not a rule but a discretion, and whether it has been exercised well for thirty years is a fact about the people who held it rather than about the arrangement. Whether the target should be alterable by the party it most affects is a constitutional question, and it is the only one at issue. The woman we met earlier who walked into the Beirut bank was not making a point about inflation. Her deposits had been fixed by a rule, in a currency pegged for more than two decades, and it was changed by the party it most affected, quietly and without a vote, when they needed the money more than she did. That is all the objection amounts to, and it needs no view about inflation at all.

¶In this context the four conditions hold over money, cleanly. The reference value is a supply schedule: a number and a rate, twenty-one million issued on a declared curve. No cleaner instance of a testable rule exists anywhere, because you can ask of any proposed block whether it obeys the schedule and the answer is yes or no, with no interpretation required and no judgement to exercise. Verification is a handful of signature and hash checks. Producing a valid block is astronomically expensive and checking one is trivial, and the gap is something like twenty orders of magnitude. A laptop in a bedroom can therefore audit a global monetary system in full and owe nobody an explanation for doing so, and it is not a technical curiosity. It is the entire reason the arrangement is constitutional rather than merely clever: the power to check is distributed to everyone, at negligible cost, permanently.

¶Somewhere tonight a student in Lagos, Nigeria, a pensioner in Arizona and an engineer in Hanoi, Vietnam, are each running a full Bitcoin node, a complete copy of the ledger that checks every block the miners produce against the protocol, on hardware that cost less than a month’s rent, and between them they hold the issuer of the world’s largest non-state money to a rule none of them wrote and not one of them could bend. They have never met. Their interests have almost nothing in common. That is why the rule holds. The mining that produces blocks has concentrated into a handful of pools, but the verification that decides whether one counts has not, and that is what holds the rule.

¶The condition about checkers who cannot be cheaply impersonated is what proof-of-work is for. The puzzles are not securing the ledger by being difficult in some general moral sense; they are making identity expensive. In a network where anybody can appear, the cheap attack is not to break the rules but to be everybody, manufacturing ten thousand participants who all agree with you. Burning electricity makes that attack cost something, by now about as much power a year as a mid-sized European country. It is also the answer to the standing objection that the arrangement is wasteful. The expense falls on whoever wants to add to the ledger, and it buys the single thing a cheaper vertical system cannot supply, which is that nobody has to be trusted. Proof-of-work is a solution to the checker condition. It is not the constitutional principle, and confusing the two has cost the argument years.

¶There is also a further governor nested inside the whole. New bitcoins that get minted go to whichever machine first solves a hard arithmetic puzzle, and the puzzles are deliberately wasteful. Those puzzles would ordinarily be solved more quickly as more machines join the contest and chips improve, and the coins would pour out faster. So roughly every two weeks the protocol retunes its own difficulty, autonomously, with no hand on the dial, making the puzzle harder in step with the network’s growing computing power, so that a block still arrives on average every ten minutes however much hardware is aimed at it. The result is a negative-feedback loop holding issuance constant as compute cheapens: a measure of the system’s own output bent back to govern its input, which is the centrifugal governor from the opening of this book, rendered in money. And the consensus rules are the execution path. An invalid transaction is not punished. It simply does not exist, because no node will write it to the ledger, which is the only place a transaction can be. There is no court, no finding and no redress, because there is no violation to remedy — it simply cannot be done. This is what enforcement looks like when it runs at the same speed as the process it governs.

¶The energy poured into the puzzles buys security: to rewrite the ledger an attacker would have to outspend the entire honest network in electricity and hardware, which makes the record costly to fake and therefore trustworthy without a trusted centralised keeper. The cap and the difficulty adjustment buy something else, which is the assurance that the issuance schedule cannot quietly be loosened to favour whoever holds the printing press. The expenditure makes the ledger expensive to corrupt. It is not a reserve standing behind each coin. Bitcoin is secured by energy, not backed by it, and you cannot take a coin to a window and redeem it for a kilowatt-hour, because the electricity went into protecting the history, not into a store beneath the price. There is a deeper symmetry, and a stroke of genius, that even the most committed Bitcoin advocates miss, because seeing it requires the idea of subsumption: it inverts the priesthood’s own physics. Effective accelerationism takes its authority from thermodynamics: the universe runs down, energy disperses, intelligence hastens the dispersal, and to accelerate is therefore to obey a law of nature. Bitcoin spends energy too, lavishly, but to the contrary end. Its expenditure is not dissipation in the service of a process that uses us. It is energy deliberately burned to hold a commitment a community has chosen to keep, against subsumption, which would otherwise dissolve it. That is purposive thermodynamics: the same physics turned from a fatalism into a tool.

¶All of which could be read as a description of a design. What makes it evidence is that the mechanism has been tested under pressure in Bitcoin’s own history: its enforcement, and the condition that keeps the rule beyond any single hand. The first is the Bitcoin block-size conflict, which ran from roughly 2015 to 2017 and which anyone who was in the industry at the time remembers as an unpleasant, sustained and expensive civil war. The proposal was technically modest — raise a parameter governing how much data a block may contain — and the coalition behind it was formidable: most of the industry’s significant firms, a majority of mining capacity at several points, a large share of the loudest public voices, and a signed pact, the New York Agreement of May 2017, among the companies that between them handled most of the world’s bitcoin. Its scheme for doubling the limit, known as SegWit2x, was called off that November, and the parameter held. And it held for a reason that has nothing to do with anybody’s virtue. The power to alter the rule was distributed among parties whose interests diverged — miners, exchanges, protocol developers, people running nodes at home, long-term holders — and no coalition could be assembled that included all of them, because each group’s gain was another’s loss. The condition about divergent interests was doing work. It was not decoration.

¶Two more episodes are older and shorter, and each isolates one part of the mechanism the block-size war showed holding as a whole. The earlier, in August 2010, was a straightforward breach: a transaction exploiting an arithmetic flaw created 184 billion bitcoin in a single block, nearly nine thousand times the supply that would ever exist. There was no argument about whether to reverse it, because there was nothing to debate. Anyone could set the block against the schedule and see that it failed the test, and within five hours a patched client had been released and the honest chain had overtaken the bad one, which was abandoned along with everything built on it. A handful of developers acted, without a vote, and nobody has ever called it illegitimate, because they were not deciding anything but enforcing a rule the network had already agreed, against a state anyone could check had violated it. That is the plainest form of the distinction everything after this depends on: a correction needs no politics when the rule is a test.

¶The later was a near miss, and it exposed the condition the other two rest on. In June 2014 a single mining pool, GHash.io, briefly held more than half of the network’s hashing power, which is the point at which one party can decide which transactions confirm and rewrite recent history at will. Nothing in the protocol prevented it, the pool announced it would cap itself below 40 per cent, miners drifted to its rivals, and the moment passed.⁷¹ The requirement that the power to alter the rule be spread among parties whose interests diverge had, for some hours, stopped being true, and what stood between the network and a single hand on the dial was that hand’s forbearance. The republican tradition has a name for a freedom that survives only at the pleasure of the party able to withdraw it: domination. The conditions are contingent achievements rather than properties of the code: a system can satisfy them one week and fail them the next, and this one came within a day of it. The near miss belongs to a network’s infancy, when participants are few and one can grow large before anyone notices, and that is the point rather than the excuse: any governor built in any domain will pass through the same period, held during it by nothing stronger than the restraint of whoever arrived first.71 GHash.io, “Official statement on 51% threat and closed round table”, 16 July 2014, pledging not to exceed 39.99 per cent of the network’s hashing power.

¶One test is still being sat, because cryptography ages. The elliptic curves that underpin today’s keys will fall to a quantum computer of sufficient size, not by brute force but by an algorithm Peter Shor published in 1994, which runs backwards the one-way sum that keeps a private key safe while its public half is visible. NIST, the American standards institute that governs the curves, has scheduled their deprecation for 2030 and their withdrawal by 2035, and their successors, lattice and hash mathematics chosen in an eight-year open contest, were published in 2024. What fails when that day comes is the signature, who may move what, and not the hash, so a ledger’s history and its consensus stand while the keys need to be swapped out. The check must therefore survive the replacement of its own primitives, because a rule that cannot renew its own locks is held not by its users but by the calendar. In a way this is crypto’s own millennium bug, counting down: only the quantum threat has no date, only a threshold, and whoever crosses it first can sign as if they were the owner of any key that has been exposed. Bitcoin is the biggest prize of all, because about a quarter of the supply sits in addresses whose keys are already exposed and whose owners are now understood to be long gone. What becomes of that reserve, whether it is frozen, burnt or simply left as a bounty to the first machine that can take it, is a question leaderless money has never had to answer before, and it is being settled now by voice and nothing else. Either outcome is evidence: a migration completed is the recipe’s second precedent, this time for changing a rule rather than holding one, and one that failed would show that voice steers a protocol less well than the recipe supposes.

¶Money is the easy case, because a supply schedule is a quantity, and a method that worked only where the rule can be written as a number would serve that one domain and little else. Most of what matters in life cannot be counted that way. Ethereum’s real contribution was to show that the rule does not have to be a number. It can be a procedure. The usual story about Ethereum is that it put more kinds of digital asset on a blockchain, which is true but fairly uninteresting. What it did was let anybody place a program on the ledger, a smart contract in the industry’s term, that holds funds and moves them according to its own code. Given the same starting point and inputs, it always reaches one result, whoever runs it. So whether a contract did what its code says becomes something anybody can check, by replaying it and comparing. The test the method needs, a rule with a yes-or-no answer, can therefore be met by a set of instructions as well as by a quantity.

¶This converts Bitcoin from a special case into the first member of a family, not of things on ledgers but of rules whose adherence anyone can check more cheaply than they can be broken. The second member of that family, arriving barely seven years after the first, was already something that could not be expressed as a quantity. In 2016 a venture fund called The DAO was assembled entirely in code on Ethereum, and within weeks it held around a seventh of all the ether then in existence, the largest crowdfunding event in history to that point. In June an attacker exploited a flaw in its contract — the code paid out before it recorded that it had, so the same withdrawal could be made to repeat — and drained about a third of the fund. What followed is the case that matters, because the response was not a patch but a decision. The Ethereum community debated for weeks and then altered the ledger itself to move the stolen ether back, and a minority who held that the code’s result was all the law there was refused and kept the unaltered chain running, so the network split in two. Both halves survive, the majority’s as Ethereum and the dissenters’ as Ethereum Classic. The decision was voice, however rough, and the dissenters who refused it left together, carrying the shared history with them, which is a different act from a single holder quietly slipping away.

¶It is not a story about villains: the alteration was legitimate by any reasonable democratic standard, proposed openly, argued over for weeks, endorsed in every informal vote available and backed by a large majority of everyone with a stake, so that the honest answer to whether it had the consent of the governed is broadly yes. That is why it belongs here. The reference value turned out to be movable by concerted action once the stakes were high enough, and the hand that moved it was a legitimate majority. The danger a designer has to fear is therefore not corruption but agreement, a large number of decent people concluding together in a crisis that a rule needs to change to adapt to a new reality, which is roughly how every failed constitution has evolved. And the claim is falsifiable, which is what keeps it from being an article of faith: the circumstances under which the mechanism does not hold have just been described, and anyone can go and look.

¶Contracting also shows the first weakening. A contracting governor holds what the code does and is silent on whether you should have trusted it, so it will enforce a predatory arrangement as faithfully as a fair one, with the same indifference; anybody who has watched crypto markets has seen people lose everything to deterministic smart contracts that executed flawlessly, in the draining of The DAO in 2016 and in every cycle since. So the governor has become more general and, in the same step, less decisive. It guarantees that the rule is followed and says nothing about whether it deserved to be. That trade — generality bought at the price of decisiveness — is the pattern of the entire argument that follows. A reader who sees the gradient here will see the third case, cognition, for what it could become.

¶The cognitive domain already has a candidate for the role, and it is worth testing first because it shows what the four conditions do when they meet a live case. On 12 September 2026 Anthropic committed, on its own, to giving independent evaluators of the kind METR represents employee-like access to its frontier models, with a contractual right to publish what they found, as part of a proposal to pace the field’s progress.⁷² OpenAI’s answer the same day was a post by Sam Altman saying it would do likewise, a statement rather than a policy. Anthropic’s step is a concession, because it hands a party outside the firm a power it did not previously have. That aside, run it through the conditions we have outlined and it satisfies none of them.72 Dario Amodei, “We Must Pace the Frontier”, darioamodei.com, 12 September 2026; Anthropic, “An update on independent evaluators”, 15 October 2026.

¶Pacing is a wish about how fast the system should move, not a question anyone could put to it and get a yes or no, and what the evaluators measure is a judgement about how a model did on a chosen set of tasks, with a margin of error wider than the result. The door is opened by the party being examined, so verification is rationed rather than cheap. There is a handful of auditors, not a population, funded from the same pool as the labs and with interests pointing the same way, while the diffuse crowd of divergent checkers is the open ecosystem that package’s rules on exporting chips and copying models are designed to keep out. And the scrutiny happens before release, convened by the lab, published afterwards. Whether it ever becomes a governor is not a question of anyone’s good faith but of where the check sits, inside the execution path or outside it.

¶The need for divergent checkers has since been half-conceded from the frontier. Meta’s reply to the pacing proposal asked for “a larger and more diverse ecosystem of evaluators”, which is the population this chapter requires, said by the party that would engage it.⁷³ A roster whose access, money and right to publish all run through the firm assessed is diverse the way the rating agencies were before 2008, several of them, each paid by the issuer of what it rated. Diversity is a property of who pays and who may publish, and of price, because a population of checkers appears only where one check costs what a hobbyist can bear. What the mechanism can hold, once it turns to reasoning, is not the answer but the conditions around it. A ledger proves whether a record was altered. It has no way to settle whether an argument is sound, since two people may reason well and reach opposite conclusions, and much of history’s most valuable work looked like error when it first appeared. The method carries over and its object changes. Any machine built to rule on correct thought would become the single authority, working at the speed of software and beyond appeal, that this book exists to prevent, so a constitution cannot promise that a cognitive system will be truthful. What it can do is stop any one party from becoming the only source able to tell you anything, by holding open the conditions under which conclusions are reached and challenged.73 Mark Zuckerberg, statement of 15 September 2026, as reported by Benzinga via Finviz.

¶Those can be tested where the content of an answer cannot. What is checkable in the cognitive domain is not whether a conclusion is correct but a set of procedural facts about how it was produced and what can be done with it afterwards. Has the system declared what it is optimising for, and could that declaration be changed without anyone noticing? Does an output arrive with the grounds on which it was reached, in a form someone other than its maker can check, which is what the sub-postmasters were never given? Is the record of your own reasoning held by you, portable and beyond anyone else’s power to erase? Does an assessor exist who does not profit from a particular answer, and does more than one system, with different objectives, remain available and able to work with the others? Is the dependency gap the capacity condition tracks widening or narrowing across a population over time? And were the acts that make up a polity’s own decisions authored by someone who demonstrably understood them? These are questions a patient, a borrower or a voter could put to the system shaping what they are told. Later they become seven tests.

¶Derek Mobley, an American over 40, applied for more than a hundred jobs at employers that screened candidates through Workday’s AI tools. Every attempt ended in a refusal or silence, a few answers arriving within the hour and others in the middle of the night. In May 2025 a federal judge in California let his age-discrimination claim go forward as a nationwide collective action. Nobody needed to know whether he was the right hire for any of those posts to see what was missing. The grounds behind each refusal, the second of those questions, were never given, and nothing in law required them. Eric Loomis, sentenced in Wisconsin in 2013 partly on a risk score whose method was a trade secret, had met the same silence a decade earlier. Like every other item on the list, it has a yes-or-no reply that someone other than the party being checked is able to settle. That is the ledger’s first test, a rule with a determinate answer, and the substance of an output never passes it. Taken together, they may resemble ordinary transparency and open-source good practice in constitutional language. They do, and the reason is now visible. Checkability is what makes constitutional enforcement possible at all, and in this domain it lands on procedure rather than on substance. The list is not a wish-list of things it would be nice for AI companies to do. It is, near enough, what in the cognitive domain can be governed, because it is what can be checked. The resemblance to transparency norms is not a weakness in the derivation. It is the derivation arriving where it had to arrive.

¶There is a small piece of evidence for the general shape of this, and it comes from ordinary commerce rather than from anybody’s politics. Firms that set autonomous agents loose on real work found much of the spend doing nothing, with the software idling in circles and calling themselves in to repair their own failures, until someone wrote down, as executable tests, what the job looked like when it was done. The industry calls these evals and has begun to treat a firm’s suite of them as among its most valuable property. Seen from here they are a form of reference value, stated as a test, applied in the execution path at machine speed, correcting an agent’s output much as the flyball held the engine to its pace. Nobody who built one was making a political statement, which is rather the point: wherever the substrate runs ungoverned, the people nearest the rubble rebuild the governor, in the medium, directly and at the tempo, without necessarily knowing the lineage of their invention. What the firm’s version cannot supply is the reference value. An eval holds whatever its owner’s commerce chooses, and ten thousand private governors, each bound to the ends of whoever built it, add up to the acceleration itself rather than to its constitution.

¶A testable rule is met, and cheap checking is partly there. Comparing what a model claims to be doing with what it does is far cheaper than training it, and cryptography is lowering that cost further, towards the point where anyone could afford it. Establishing that an answer’s stated grounds are real is easy in some cases and hard in others, and measuring a population’s growing dependence is expensive in the way a census is: ordinary work for an institution, not an unsolved problem. Checkers who cannot be cheaply impersonated, with divergent interests, are the challenge. In money, identity is made expensive by burning energy, and the primitive is well understood and in daily use. In cognition there is no equivalent primitive in general use, and the checkers who would matter — independent assessors, alternative model providers, auditors of long-run capability decline — are few in number today, expensively capitalised, and therefore easy to capture. Structural plurality is not a market preference in this argument. It is a constitutional requirement, and it is the condition most likely to struggle.

¶Who pays for them decides whether they exist. One of the most consequential outside checks of recent years cost seventy thousand dollars. In 2013 engineers at West Virginia University, on a grant from the International Council on Clean Transportation, a small non-profit, drove three diesel cars around California fitted with portable measuring equipment, and found that two Volkswagens emitted nitrogen oxides at up to thirty-five times the legal limit on the road while passing every official test.⁷⁴ Their software could tell when it was on an emissions rig and cleaned up the exhaust only then, a behaviour evaluators of frontier models now watch for. Within two years the US Environmental Protection Agency had served a notice of violation, and some eleven million vehicles worldwide turned out to run the same code. None of it needed the manufacturer’s permission, and nobody with a stake in the result was paying. The assessors who matter are few because assessment is dear. Bring the cost of a single check down to what an amateur can carry, as the Bitcoin full node did for money, and the population emerges on its own. The cryptography that lets a checker confirm a computation without repeating it is the same instrument seen from the other side, and the transcript of a failure, once reproducible, is cheap to verify however hard it was to find.74 Gregory Thompson, Daniel Carder, Marc Besch, Arvind Thiruvengadam and Hemanth Kappanna, “In-Use Emissions Testing of Light-Duty Diesel Vehicles in the United States”, final report to the International Council on Clean Transportation, Center for Alternative Fuels, Engines and Emissions, West Virginia University, May 2014.

¶What remains is the purse, and the requirement on it is the one that applies to any auditor: whoever pays must not be the one who picks, and the result must publish whether or not the subject consents. None of that stops the laboratories paying. They can fund the whole population between them, provided four things are kept apart that a bounty programme keeps together: they do not choose the targets, they do not judge the findings, the payout is triggered by a transcript anyone can replay rather than by anyone’s verdict, and the purse cannot be withdrawn by the party it is spent on. Whether that last comes by levy or by an agreement made irrevocable is for legislators and laboratories to settle, and this book prefers the former, for the same reason that every publicly listed company funds an audit its management is barred from appointing. A network rewarded for discovery and paid on proof has no stake in what it turns up, which is what makes the miner the right model. How that is engineered is for founders to work out, and for the market to incentivise.

¶Nor must every act of cognition be checked, any more than every taxpayer needs auditing. A sample drawn at random from what the systems did, across providers and unannounced, holds all of the output to the reference value at a fraction of the cost. What the sample cannot do is settle a single case, so it measures the setting and its drift rather than any one outcome, which is the scale at which this section has said the capacity is lost. Three things keep it honest. The population sampled is fixed before the draw, so that the party measured cannot choose its good days. The draw is made by someone else. And the records drawn belong to the people who chose to submit them, which portable memory, one of the tests proposed later, makes possible and which no laboratory could grant on their behalf. Where a check sits is a design question with real content, and not mine to settle. Which checks can be moved into the execution path, at what cost in latency, and which have to sit alongside it and accept being slower, is the sort of question founders and the market are better placed to answer, and I should decline to. What the book owes you is the criterion. What to build against the criterion, and at which parameters, is engineering, and it will be revised as the parts improve, which is the correct relationship between a constitution and an implementation.

¶Nick Land’s book on Bitcoin, Crypto-Current, makes a claim close to mine. He argued that Bitcoin has turned rigorously constitutional governance, a problem the old liberal order could never solve, into a technical capability under automatic propulsion, and that the industrialisation of government follows, with regimes becoming, potentially, business outputs. Land reached this chapter’s thesis before I did. Constitutional government, understood as a rule that binds without an institution standing over it, stopped being a problem of political philosophy and became a thing that can be built. He is right. The disagreement between us is now much sharper than it would otherwise be, because it is no longer about whether the mechanism exists. It is about two things only: what reference value the mechanism should hold, and whether voice survives.

¶On the reference value, nothing in the mechanism selects one. The point made at the start is this: a governor enforces and does not decide. Land’s constitution is proprietary and disciplined by exit: a regime is a product, its customers may leave if they dislike it, and that leaving is all the say they have. Mine is a commons, and the reference value it holds is that no party may wield uncheckable power over the rest, which is the non-domination this book has been building towards. Both are reference values. The mechanism will hold either with perfect indifference. So the choice between them is not a technical question at all, and no amount of engineering will settle it — which is a point in favour of my argument and against the tendency, common on both sides of this quarrel, to think the technology has a politics of its own. On voice: a regime that is a business output has customers, and they may leave but may not speak. That is not an incidental feature of Land’s position. He states it as a virtue, and it is the point at which his argument and mine become incompatible rather than merely differently emphasised.

¶Land also posed the obvious question about proof-of-work — could all that computation not be doing something useful? — and answered it in the negative, on principle, holding that the perfect pointlessness of the generation procedure is a feature rather than a bug. The reasoning is that if the work had value elsewhere the cost signal would be corrupted, since the price of an attack would be discounted by whatever the attacker could earn from the computation, and the security would degrade accordingly. But Land is objecting to one way of meeting the need for checkers who cannot be cheaply impersonated and whose interests diverge. Proof-of-work makes identity expensive by making work meaningless. Any other mechanism that makes identity expensive and keeps interests divergent satisfies the same condition equally well, and would do so whether or not the expense took the form of meaningless computation. So his argument establishes something about proof-of-work but nothing about governors generally. That is the artefact-versus-recipe confusion this chapter exists to correct. He has mistaken one implementation of one condition for the constitutional principle, which is the same mistake made, from the opposite direction, by everyone who thinks the lesson of Bitcoin is that you should own your keys. Land understood the thermodynamics of Bitcoin perfectly well. What he did not do was centre the energy dynamic, since in his account thermodynamics enters largely as an analogy for irreversibility rather than as its substance. The narrow claim is that compute as a contested budget — the question of who decides what the world’s computation is spent on, treated as a political matter with a distribution and a set of claimants — is not in his account. Where compute appears there, it is getting cheaper, which is a fact about supply rather than a question about allocation. That is the gap this book works within.

¶There is a gradient running through the three cases, and it is in how much of the reference value the governor itself supplies. In money it supplies all of it: the target is a number, twenty-one million, and the governor makes that figure true. In contracting it supplies the rule but not the result: the target is that whatever the code specifies will happen decided by whoever wrote it. In cognition it supplies neither, because a governor that fixed what a model is for or what a person should think would be teleological subsumption with a constitution. What it holds a system to is not a goal but three facts about purposes: that it has declared one, on record, that the declaration can be checked against what it does, and that the people it influences can change it should they wish. The governor guarantees the question, never the answer. It weakens along the series, which follows from the trade-off named in the contracting section, generality bought at the price of decisiveness, and by the third case, cognition, what is guaranteed is thin indeed: not that you will think well, not even that the systems will serve you, only that the means of contesting what they are for remain in existence.

¶The reply to the objection that this is too thin to be worth having is a claim about what constitutions are for. No constitution has ever guaranteed outcomes. What a constitution does is keep outcomes contestable. Free speech does not ensure anybody speaks well, but it makes silencing answerable. The vote does not guarantee a wise choice, but it renders power removable. Due process does not produce correct verdicts, but it makes wrong ones challengeable by the person they fall on. Every constitutional guarantee has this thin, procedural shape, and every one of them has turned out to be worth more than the substantive guarantees offered instead. So what is being proposed for cognition is a constitutional guarantee of the ordinary kind, in a domain that at present has none at all. It has to be constitutional rather than left to the market, which is where David Sacks would leave it, for the reason the capacity condition gave: trade disciplines bad actors through the judgement of the people doing business, but it cannot protect the faculty it depends on. The requirement therefore belongs in the rules every polity shares and not in someone’s terms of service. This is also the point at which the transfer argument and the capacity condition meet. Both end in the same result: the constitution creates the conditions and cannot compel their use. A polity that has the means of contesting a thing and declines to use them has not been failed by its constitution. It has made a choice, which is all a constitution can preserve for anyone.

¶There is one place this argument will be fought, and it is not a philosophical objection. The condition that the checkers cannot be cheaply impersonated may be challenging in cognition specifically, and the reason is not money. A Bitcoin full node needs nobody’s permission: the ledger is public and the rules are in the open. Checking a frontier model depends on access to its maker controls: the weights are private, serious evaluation needs entry the laboratory can grant or withdraw at will, and most people able to do the work are former employees of the firms under scrutiny, sharing their trade, their outlook and their future employers. A purse does not cure that. Only a rule about access does, and the tests proposed later supply one. So the bet is empirical and it is about structure rather than price: whether the capability to check can be held by parties with different interests, or it stays inside the building whatever it costs. Should the tests be adopted, outside checkers would gain entry by right, not by favour, and the wager could then be lost on its own terms. If they keep missing what the makers’ own teams catch, or verifying a model becomes as dear as building one, the condition of divergent checkers fails in cognition and the method does not carry over.

¶So: Bitcoin holds a single reference value. It re-imposed scarcity on a substrate that had abolished it, and it proved that a rule can be written in code and held at the speed of what it constrains — the machine — with no institution above it and no hand on the tiller. It is not a model for a society, nor is it a currency I am urging anybody to adopt. What it demonstrates is a method that transfers under four stated conditions which I have now checked in three domains and found to hold completely in one, partially in the second, and unevenly and with one serious doubt in the third. Scarcity is not the only thing subsumption dissolves and money is not the only domain that needs a governor. Provenance, personhood, cognition and the conditions of plural authorship all meet the same solvent and will want a governor cut to their own shape, and each will have to be taken through the same four tests. It is proposed anyway because the doubt is empirical rather than principled, because the domain at present has no governor of any kind, and because a method that names the condition on which it fails is worth more than the alternatives on offer, which identify none.

¶None of this is lost on the priesthood, and it is something my venture capital firm and I share with them. The same venture treasuries that pour capital into the frontier labs also often fund these same primitives — the distributed ledgers, the tokens, the cryptographic rails — and they are right to, because the tools are real and powerful. What differs is what the primitives are bound to. In their assembly the cryptography is plumbing for the machine economy, a way to let autonomous systems transact and settle without human friction: incentives with no reference value above them, the pure accelerant this book described earlier. A heap of ledgers and tokens is no more a governor than a pile of flyballs and linkages, until it is bound to a reference value, just as Watt’s loose parts became one only once they were arranged to hold a speed. The priesthood reached for these and wired them into the deletion. The same parts, wired instead to a reference value a community has deliberately chosen, become a governor. Same primitives, opposite constitution. The mechanism holds whatever it is given and does not care what that is. So the next thing the argument owes is the purpose the reference value serves: what, in the end, all these governors are for, and by what right anybody sets it. That is not an engineering question and no protocol will settle it. It is the oldest political question there is, arriving in a new place.

Chapter 11

On Sovereignty: The Trinity

“So this is the formalist manifesto: that the US is just a corporation. It is not a mystic trust consigned to us by the generations.”

— Curtis Yarvin, “A Formalist Manifesto”, 2007

¶When Curtis Yarvin above calls the United States “just a corporation”, he is making an argument about sovereignty, and that word has been doing quiet work throughout the book, carrying four conflicting meanings often read as one. The most cited belongs to the creed of the sovereign individual, for which it is a personal exit: the capacity of someone cognitively and economically able, and mobile enough, to take themselves and their assets beyond a jurisdiction’s claims. That is a private good, held against everyone and available in proportion to one’s means. For Land it is economic, or catallactic, a term he borrowed from Hayek for a kind of order that comes out of buying and selling, and on his account no person really holds it. It sits wherever capital happens to be flowing, in the service of an intelligence that arises from markets and machines that keeps making more of itself, and even the sovereign corporations of his patchwork only have it on loan. For Yarvin it cannot be split or shared, because someone always has the final say, and any system that claims to spread authority around is only hiding who that really is. He concludes that a state should be run like a company, under a proprietor who admits as much, and that divided government is just a pretence. This book instead uses the word politically, for a people’s standing to set the terms they live under and to change them, shared among them by design and never held by any one party. The other three sit closer to one another than any does to this one, because each treats sovereignty as property and differs only over whose name is on the deed, whether a person over their own affairs, the market or a ruler. On that view the only remedies are the ones any owner has, to sell, refuse to buy or leave, and nothing in that arrangement lets one holder compel an account from another. That is where the sovereign-individual tradition ends up, at Land’s formula of no voice and free exit. Its thinkers were not indifferent to being heard, but a conception built on ownership has nowhere to put an objection, so leaving is the only move it allows.

¶Shared sovereignty still needs something to anchor it. Bitcoin held one invariant perfectly: scarcity, re-imposed on a substrate that would otherwise have dissolved it. That is the job of a governor, to pin a system to a chosen setting, and here it was elegantly the number twenty-one million. Nobody had to tell the protocol why. Its makers, working in the cypherpunk tradition, had already decided that a currency no state could debase was a form of liberty, and the cap carried that judgement. Elsewhere there is no such shortcut, no figure that settles the question for you. Scarcity protects money because a currency keeps its worth only while its supply is limited, but no cap shields a person’s identity, attention or thought, and subsumption is wearing all three down at once. So every new instrument needs someone to choose its target. Letting each domain pick its own does no harm in itself, and the aims may even balance one another, but only while somebody watches what they add up to. An identity scheme built for security demands that everyone be verified, while a payments network designed for compliance makes each purchase traceable, and each is reasonable on its own terms. Combined and unwatched, they would hand one power a dangerously complete picture of a life. That oversight calls for a standard to judge by, so before the method can travel, the people using it should decide what, in the end, their tools are all collectively for — the one value the governance stack exists to protect. Here I agree with Marc Andreessen that freedom comes first, since nothing else can be chosen, revised or answered for without it. I do not mean it as he usually does, though, asking only that nobody interferes with you, but rather in the sense of not living at the mercy of someone else’s will, what republican political theorists call non-domination. It is what the slave with a kind master lacked, as we saw when we mapped the three roads to totalisation earlier, since he could be left alone for years and still never be free. This is the spine of all that follows, and every governor proposed later answers to it.

¶Everything from here rests on one assumption: that a person’s standing to weigh, choose and answer for their life is worth keeping for its own sake. Almost nobody this book addresses is likely to deny it. The priesthood certainly doesn’t. Andreessen’s own manifesto celebrates technology as the crowning work of human ambition, and when Zuckerberg answered Anthropic’s pacing proposal he rested his case on people refusing any agent that did not serve them. Whilst some strands of posthumanists may think our species is a net cost to the planet and would welcome its retirement for everyone else, the open question is what keeps agency alive. So why freedom? Health and prosperity are pretty important to most people, as is the absence of pain. A capable enough system might deliver all three in quantities no free society has yet sufficiently managed to achieve. If the trade-off is a longer, richer and easier existence in exchange for a life that is less free, most reasonable people would be inclined to give it serious consideration. The answer is that almost all the ways we run our lives together depend on some form of freedom, and losing it does more damage than first appears. We make promises and hold each other to them. We enter contracts. We consent. We praise, blame, forgive, apologise, resign, prosecute and pardon. Every one of these assumes that somebody made a choice and is there to formally answer for it. That connection between freedom and moral responsibility is why cognitive subsumption, once it puts us at the mercy of whoever runs the system, is more serious than the loss of a job or a skill. What it outsources is the act of weighing a decision itself, the faculty by which a person arrives at a view they can be held to. A country can shed a great many capabilities and still remain a society of authors, but it cannot lose this, because every way it can hold people to account relies on it. The trade-off we began with therefore undoes itself. Consent is on that list, so whoever bargains away the power to refuse has signed the last agreement open to them, and the children who will live with the outcome were never asked.

¶A system that can predict your next choice can also make it for you, and it never has to interfere with you to do so. A recommendation engine knows what you will watch. A model drafts the reply you were about to write, and a pricing algorithm has worked out what you will pay. None of them touches you, yet each narrows what you do. That is why freedom here means something modest and practical. The shape of your life, for better or worse, should trace back to your own decisions, even though much of it is inherited, constrained or luck. At its core is the ability to weigh a judgement, to refuse it and, above all, to do what nobody saw coming. Progress depends on that last one, from the innovator who builds what the market never knew it needed to the scientist whose breakthrough no one planned. Friedrich Hayek believed a society that can anticipate and pre-empt its members’ choices is poorer in liberty and, for the same reason, in invention. Hannah Arendt’s two conditions, met earlier, describe what keeps it alive: plurality, the many who see the world from different places, and natality, each newcomer’s power to begin something unforeseen. Where both are in place, a society does not have to decide how anyone should live, because everyone stays free to hold, change and argue over their own idea of a good life. So all we need to ask of the governance stack is to keep choosing open, and in many hands.

¶So if we agree freedom is a reasonable minimal reference value for society, the next question is which kind. Political philosophy uses the word for at least three things, and the priesthood often quietly trades on the ambiguity between them. The first is negative liberty, Berlin’s freedom from interference, met at the start of this book. Its virtue is that it is clear, testable and hard to abuse, because it can be easily felt and monitored. Its weakness is the one the republican tradition pressed: it only counts acts, so it cannot see a standing power not currently being abused. By its measure the slave with a kind master is free, as are the tenants of a benevolent landlord and the West Virginia miner of a century ago, paid in scrip he could spend only at the company store. By that logic, so are all of us whose reasoning AI quietly steers, since it has never once told us what to think. Every one of them, us included, lives on sufferance, yet because nobody is interfering, negative liberty calls them free. The modern version is quieter. A developer who has spent five years building a business on one laboratory’s model is a fair example. The lab has not cut off his access, raised his price or launched a product to compete with his. He is, by the negative test of freedom, free, and yet he must arrange his life around the knowledge that any of the three could come within a week with no recourse. Dependence of this kind is, in outward form, indistinguishable from liberty, which is what makes it hard to legislate against and easy to live inside without noticing, the way Ellul’s fish is unaware of the sea.

¶The second is positive liberty, freedom as capacity, which the starving man on Berlin’s empty plain lacks and which Amartya Sen taught development economists to measure, as we saw at the start of this book. It is the sense in which two centuries of material progress widened human lives. Its weakness is the one Berlin warned against, with the Soviet state in mind: once freedom means having certain capacities, someone has to choose which ones matter, and whoever decides gets to force people towards their own particular form of liberation. Every regime that has claimed to know its subjects’ true interests better than they did, from the Jacobins to the Bolsheviks, used this argument, so any governor we propose has to be built so that it can never be bent to the same end. The third is not a philosopher’s account, but the one the priesthood’s doctrine runs on: freedom as optionality. More choices equals more freedom, and a system that multiplies what is on offer enlarges liberty by that fact alone. It has the great advantage of being measurable and yet the fatal defect of being silent on what matters most, because it never asks who drew up the list in front of you, to what end, or whether you could even find out. Our preferences are formed by what we have been offered, and nobody craves a fruit they have never tasted, so the hand that sets the options shapes the wants as well. By this test, twenty choices picked by someone whose interests run against yours, in a way you cannot see, count as more freedom than five you found for yourself.

¶If freedom matters for these reasons, what would it take to secure it in practice? Here again the republican tradition supplies the test, since a master’s forbearance can be revoked at any moment and answers to nobody. An AI model that decides what you see, read and consider never lays a hand on you, but interference, in the republican account, has always meant more than just violence. Pettit lists manipulation among the forms it might take: fixing what a person gets to consider, or shaping their beliefs and desires by means that bypass their own reasoning. That is very nearly what an AI does as a matter of course when it stands between a population and what it reads, searches, drafts and decides. Choosing which considerations reach you is fixing the agenda. Shaping what you believe without ever appealing to your judgement is the manipulation Pettit meant, however good the intentions or accurate the answers. Neither requires a lie, or a desire to harm. But both still count as interference on Pettit’s terms, because the objective behind them sits inside a model you cannot inspect or contest, and so in this sense they are arbitrary. That is domination by definition rather than by analogy.

¶The anarcho-capitalist reading treats negative liberty as if the interfering party were always the state, though Mill, Berlin and Pettit, as we saw with the cypherpunks, never did. Private power that answers to nobody is a failure of freedom even on negative liberty’s own terms, and it falls with full weight on the handful of firms that now hold the frontier of cognition, whose forbearance is the modern master’s. There is, however, also a failure that does not look like tyranny at all. A polity can keep every formal instrument of contestation and yet lose the ability to change the outcome. The vote still happens, the courts still sit, the press is not censored and the right of appeal survives. The consultation still runs and the responses are published. By every institutional measure, voice is intact. But if the options from which a people choose have already been narrowed by a power they cannot inspect, then what survives is the form without the substance. You are still free to object. There is simply nothing left to contest, because the question was settled before you were ever asked, by a process you had no way of seeing and no standing to challenge. Call this a hollow voice, a failure distinct from censorship, capture or apathy: the quiet replacement of the real thing with an imitation of it.

¶Republican thought has never been satisfied by a remedy that lives only on paper. It requires that power can be called to account in practice. Two conditions follow from this for our work. The first is political, answerability, a channel through which power can be challenged. The second, capacity, is developmental, a population still able to form a view of its own and to imagine options nobody put in front of it. That is different from poverty, where the choices are few and everyone can see it. In a hollow voice they exist, or could, but someone else has taken them out of view, which makes it a question of power rather than of means. Neither of the two is sufficient alone, so they have to be held together. If keeping people able to think for themselves is the constitution’s job, then the charter needs a real measure of that ability, one precise enough that a country could fail it, and the seven tests proposed later in this book are built to be that. Answerability depends on something it cannot supply for itself. To contest what a system is optimising for you have to be able to evaluate it, and that ability is the one this book has shown being worn down. So the remedy presupposes the capacity whose loss it is meant to prevent. It works only as long as it is not needed, and stops the moment it becomes necessary. A right you can no longer exercise is a formality, worse than nothing because it makes an arrangement look constitutional while leaving it unconstrained. You can lay on a bus to take addicts to work, but you cannot give them the capacity to get on it, and a simpler complaints procedure, on its own, only lowers the bar while the skill to clear it keeps falling. That is why the two conditions stand together rather than one inside the other. Answerability rests on non-domination, which was never simply being left alone. A tax authority interferes with you constantly, yet where it can be questioned, appealed and removed you are not at its mercy. The point is not that a model should stay out of your thinking, since helping you reason is what it is for, but that whoever set its objective can be found and made to respond when you contest it.

¶Capacity covers what answerability cannot reach. You can consent to a system, inspect it, challenge its objective and get a reply, every right intact, while it quietly makes you less able to do any of those things. Nobody is your master and no hand holds power over you, yet the faculty all of that depends on is being drawn down by a process you agreed to, at a rate first measured in 2025, and then only as the by-product of a study designed for something else.⁷⁵ That is a failure of development rather than domination, and it needs its own condition because answerability cannot see it. Opting out does not help. The harm is positional, in the sense met with the refusers, since what your judgement is worth depends on everyone else’s, so someone who declines the tools while colleagues, rivals and children all use them simply falls behind. Because the damage is structural, the remedy should be constitutional rather than advice to the user. It should be a property of the system, testable from outside, and what it measures is dependency, in the sense the criteria gave it.75 Krzysztof Budzyń and colleagues, “Endoscopist Deskilling Risk after Exposure to Artificial Intelligence in Colonoscopy”, Lancet Gastroenterology & Hepatology, August 2025.

¶Non-domination is a standing you hold in relation to power, and is maintained by being used. A right nobody exercises decays into a permission, because the party it constrains gradually discovers that nothing happens when it misbehaves, and the lesson compounds. Contestation is the system’s error signal. Where nobody objects, the party holding power stops learning where it has gone wrong, and its conduct drifts. None of this can be seen from inside, because the signal that would have revealed it has stopped running. A governance stack that guarantees rights to a population that never uses them has only automated the appearance of freedom, a piece of theatre crypto has already staged. Much of what has passed for on-chain democracy worked this way, with proposals put to token holders at short notice, turnout often in the low single digits and outcomes settled in advance by a handful of large wallets, so that everyone held a ballot and almost no one had a say. The lesson for those designing such systems is unglamorous: contestation has to be made cheap, ordinary and habitual rather than heroic. If challenging a decision requires a lawyer, several months and a high tolerance for personal cost, the right exists and the standing does not. If it requires a fork, a migration and the abandonment of everything you have accumulated, likewise. Most of what a good constitution does is lower the cost of using rights rather than declare them. A captured system rarely removes them. It raises that cost until they go unexercised and, in time, unremembered.

¶Perhaps the atrophy of our own judgement, as we hand more of our thinking to machines, need not be terminal. A community that stepped back from these tools, as some religious groups have held apart from the modern world before, could keep the faculties the rest of us lose and return them to us later, as custodians, just as Latin Europe recovered the Greek inheritance from Arabic scholars who had preserved it. That holds on only one condition: that such a group can survive next to a far mightier neighbour, as the United States eventually came to tolerate the Mormons of Utah. But this requires goodwill, and indifference lasts only until the enclave stops seeming a curiosity and starts to look like a competitor or inconvenience. The dominant party need not be a state. It could be a superintelligence, and Elon Musk, introducing xAI in July 2023, mused that to such a mind humanity would be at best a curiosity, so our hope as a species lies in being interesting enough to keep around. Either way the enclave lives on the forbearance of a power it cannot contest, which is the position of the slave with a kind master again, who is left alone for now but could be interfered with at any moment without recourse. The tolerated community, whether the Mormons or humanity as a whole, finds itself in the same predicament.

¶The enclave fails on its own terms as well, because the reservoir was never meant to be a library. Manuscripts are the easy part, since hard drives survive fires, empires and centuries of neglect. What the reservoir has to carry is agency itself, which is not an object to be stored and handed on later but a standing in relation to power. A community that lives by permission may keep every practice, craft, language and book intact and still lack the one thing recovery needs: the ability to set its own ends against a force that would rather it did not. The Arabic-speaking world did not preserve Aristotle at Latin Europe’s sufferance. It preserved him as the working property of civilisations setting their own course, which is why what came back was a living tradition of argument rather than a crate of documents. A dominated reservoir can pass on the memory of agency, the texts and rituals of self-government, but not the power to act on them, because it has none to give. That is terminal subsumption again: everything remains except the ability to use it, so any recovery would happen on terms someone else dictates. Such a community could stand undominated beside a far stronger neighbour only if both were bound by a rule neither could change alone, which is the work of a governor.

¶Then there is the future in which no one is ruled at all, only no longer needed. There is no overseer, no boot on the neck and no command to obey. People are simply surplus to requirements, routed around by the market, priced out and made economically irrelevant by systems that bear them no ill will and give them no thought. Being unneeded is not the same as being mastered, but it becomes domination the moment your livelihood depends on a party you cannot contest, which is the position of the tolerated enclave. The harm lies less in losing the job than in having no recourse when it happens. Someone laid off by a firm in a functioning labour market — with savings, a union, benefits paid by rule rather than favour, an appeal to a body obliged to answer and other employers ready to hire — has suffered a serious loss but has not been dominated. Take those away, make the decision-maker unreachable, the review a form nobody reads and the support conditional on an authority beyond question, and the economic fact is unchanged while the political one is not. Gig workers know the difference well. In April 2021 a court in Amsterdam, the Netherlands, ordered Uber to reinstate six drivers — five in Britain and one in Portugal — whom it had dismissed by an automated process, and to pay them compensation. None had been told why, the notice said it was final, and asking again returned an identical message. Losing the work was an ordinary misfortune. What turned it into domination was that no one at home could reverse it, and the drivers had to look abroad for a court that could overrule Uber. The worker and the enclave have nothing else in common, but each depends on a power it cannot contest, and one account covers them both.

¶There is a way of deciding between the worlds this book has described. The American philosopher John Rawls asked what principles people would choose for a society before knowing who they would be in it. Behind what he called a veil of ignorance, no one can tell whether they will be born rich or poor, gifted or ordinary, healthy or sick, so nobody can rig the rules in their own favour, and self-interest turns into fairness. Rawls hid the generation too, and drew from it a duty to save for those who come after. Stretch the horizon and you might arrive in the generation that builds the substrate, the one that inherits it, or the one three hundred years downstream living inside whatever it became. The answer that survives is thinner than equality and harder to argue with. Whoever you turn out to be, and whenever you arrive, there should be a power you can contest. It need not treat you well, which cannot be specified in advance. But it has to answer, to somebody, by a route that does not depend on its own goodwill, and that is what none of the three roads supplies. The present may bind the future, as every constitution does, in that each generation lives under inherited laws until it changes them by an agreed procedure, but nothing should be placed beyond amendment except the capacity condition, without which no later revision is possible. Whatever is set up now must leave those who come after able to challenge it, and a doctrine that writes them off in advance, on the strength of a view of deep time they had no part in choosing, fails before it starts. The laboratories already apply this test to one another, since none would accept a rival it had no way to check. Each eyes the others with suspicion over which might capture the state first, and Musk spent years in court with OpenAI over who should control the company, until a jury in May 2026 ruled that he had sued too late. Gambling on the average result rather than insuring against disaster makes sense only where the damage can be recovered, and here it cannot, because the worst outcome is the permanent loss of the standing to change it, and the faculty by which compensation would be claimed is itself gone. That turns a question of philosophy into one of engineering. If no party may ever hold a power beyond contest, the arrangement has to be constructed so that none can dominate.

¶The cypherpunks failed not because their cryptography was wrong but because their politics shipped one part of the answer and called it the whole. They built tools that defended the individual against the state and left it exposed to everything else, to the private concentrations of the whale and the protocol owner. A single force, run on an ungoverned substrate, gathers power back into a few hands, and crypto’s first decade is all the proof needed. So the reference value cannot be the rule of any one force. It has to be an order in which no party is sovereign in itself, held in balance by the governor. Their technology went further than their politics. The forces that make a society free are old, and they have always pulled against one another. Wherever innovation has dictated the terms, the market has usually come out on top, which is the road to techno-capitalism this book has followed. What the stack the cypherpunks set in motion adds is the means to hold that tension rather than settle it in favour of one. For the first time each force can work without a gatekeeper able to capture it, and be limited by rules written into the systems it runs on. Self-custody removes the custodian, while a shared ledger binds every account, however large, to rules nobody can rewrite alone. Permissionless exchange lets anyone enter the field, while open code and the right to fork stop any platform from fencing it off. A decentralised autonomous organisation, or DAO, lets people pool funds across borders, while the freedom to leave, at a known price, means it cannot trap them. Earlier restraints, from the separated branches of a constitution to the competition authority, rested on institutions the powers they policed could co-opt or outlast. These primitives write the limit into the protocol, so the order no longer depends on the goodwill of the strongest.

¶I will call these three forces — the Individual, the Market and the Collective — and the pull each exerts on the other two, the Cypherpunk Trinity. Each is generative in its own right. The Individual is the person with keys, able to hold property, identity and speech that no authority grants and none can revoke, with an exit always in their back pocket, and from that standing comes initiative, what Arendt called natality, the power to begin something nobody foresaw. The Market is the open field of voluntary exchange, where strangers transact directly, without an intermediary but under rules no single trader can rewrite, and it yields discovery, the knowledge no planner holds that Hayek saw prices gather. The Collective is the polity and the commons, the body that acts together and keeps what belongs to all of it, and it makes possible provision and voice, the things no one can secure alone. A society that kept a single force would not be purer, only poorer, and it would hand that survivor the power to decide what the rest were worth. Like Aristotle’s virtues, each is excellent in measure and destructive in excess. Left to itself every force presses outward until it totalises, singly or fused with another, and becomes one of the roads to the unfree world traced at the start of this book.

¶Of the three, the Collective is the force most in need of definition, because it is the one the cypherpunks never articulated and the one nothing in this argument can do without. It runs as a spectrum, and the axis that orders it is not size or formality but the cost of leaving. A DAO written entirely in code has a constitution more elaborate than most statutes, and yet nobody mistakes it for a government. Walking out of a book club costs nothing. A union, a church or a DAO prices the exit — a notice period, a social cost, a forfeited stake — and the toll is known upfront and finite. Priced fairly, that toll does useful work, keeping members at the table until a quarrel is settled, provided the door stays open. A network state, a DAO that has bought land and seeks recognition, is cheap to join, but each acre, passport and treaty it gains makes the door heavier. A platform charges more than most people will knowingly pay, holding a livelihood hostage to network effects and accumulated data. And the nation-state demands the most a person has: borders, exit taxes, renunciation fees, the body itself the last thing to be carried out. On this axis the state is simply an association taken to its limit, the collective whose exits closed, and which then acquired territory and the monopoly on violence because nobody could get out to contest it. I have never been against the state as such. The enemy instead is the total state, sitting at the top of the exit-cost curve, where membership stops being a choice and becomes a condition. Its fully automated form is where the state road ends, and any association or platform can climb towards it. The firms met on the market road were collectives too, and climbed that gradient faster than any government, with no public accountability attached to the power they gathered on the way up. Cyberspace’s job on this axis is to drag the state back down the curve, so that wherever leaving is possible again, it has to compete, on taxes, liberties and law, like any other collective that wants to keep its members.

¶The Individual runs along a spectrum too, and it is the harder one to see, because its axis is not the cost of leaving but how far a person’s power outruns what they are required to answer for. At one end stands the citizen, whose property and speech are theirs and whose choices touch mostly their own life, a position worth defending for its own sake. Next comes the sovereign individual, rich enough to carry wealth, skills and family anywhere. Here the problem is less a technique than the creed met on the road to exit. It rests on a fallacy this book has already taken apart. If the advance of technology obeys a law of physics, the system that delivers it is natural too, success within it is deserved, and those who achieve it become agents of progress and therefore of freedom itself. Those who hold this view do not claim that such a person is morally better than anyone else, only that the order of things has certified what they won, so nobody has standing to question how they use it. Marc Andreessen said in a March 2026 interview that he practises introspection as little as possible. The creed explains why he can afford not to: if present success is natural, future decisions need no examining, and whoever makes them stands above ordinary politics and the voice of the people they affect.

¶Further along stands the oligarch, one of a few whose wealth and indispensability buy a private channel to ministers, though he still answers to rivals and needs the state he has captured. Taken to its extreme, that creed crowns the autarch. The autarch met earlier — whose name should not be confused with autarky, the self-sufficiency of a closed economy — would sign the sentence Peter Thiel wrote in 2009: “I no longer believe that freedom and democracy are compatible.”⁷⁶ He treats the state as a supplier he may drop, and moves between two postures without settling in either. In the first he leaves to found a polity in his own image. Thiel funded the Seasteading Institute in 2008 and became a New Zealand citizen after twelve days in the country. Starlink’s terms of service declared Mars a free planet in 2020 beyond the authority of any Earth-based government, and Musk has floated loans for settlers who cannot afford the fare, to be repaid by work on arrival. In May 2025 the launch site at Boca Chica, in Texas, voted itself into a city called Starbase, its electorate largely SpaceX employees. In the second posture he stays and takes hold of the state from within. The office Musk ran in 2025 reached into federal payrolls, contracts and data systems, among them agencies that oversaw his own companies. The alternation is not inconsistency, because the possibility of leaving is what makes his presence count. What marks that end is not success, which this book is all for, but the loss of anyone able to say no. Command over others at any meaningful scale always needed other hands, staff to run it, soldiers to enforce it, workers to make it, and each of them could down tools, which capped the climb. Subsumption removes that limit, since a single owner directing swarms of agents needs nobody’s consent to act like a ministry. Here the kind master arrives in its most complete form: a ruler who may well be generous, answering to nobody, over people who can neither leave nor refuse.76 Thiel, “The Education of a Libertarian”, Cato Unbound, 13 April 2009.

¶Hirschman explains why the alternation works. A credible threat to leave makes any protest louder, and its force turns on two quantities: how cheaply a person can get out, and how much the state loses by his going. For the richest both run high. Departure costs them a flight and a residency application. The treasury, meanwhile, forfeits a tax base, an employer and often a capability nobody else supplies, so the warning rarely needs saying aloud. Scholars after Charles Lindblom call this the structural power of capital, the discipline a government accepts from the mere possibility of disinvestment. For a teacher in Ohio or a farmer near Nairobi, in Kenya, the arithmetic runs the other way. Moving would cost either of them family, language and licence, the government would barely register the loss, and a grievance counts for little however just it is. Call the difference the exit premium, the influence a man is paid for keeping his door ajar. It holds only while somewhere will have him. Russia’s richest men bought the state’s best assets in the loans-for-shares auctions of 1995 and 1996, were told after 2000 to keep their fortunes and stay out of politics, and found in 2022 that sanctions had shut the London door from the far side.

¶The balance is held pairwise, each force checking the next, on the principle James Madison set down in 1788 in The Federalist No. 51, of giving each power “the necessary constitutional means and personal motives to resist encroachments of the others”. Give the Individual a hard floor the Collective cannot breach — keys no authority can confiscate, a wallet no government can freeze, speech that nothing silences at the source — and you have the classical guard, the one the cypherpunks got right, against the totalitarian state and any majority or stronger party that would override a person. But a person cut off from exchange is not freer, only more isolated, which is why the Market checks the Individual: to transact is to need a counterparty and to accept the trade-offs that brings, so the openness that lets you exit also obliges you to bargain. That is the guard against autarky. It breaks down where one party’s bargaining power dwarfs the other’s, as a crypto whale’s does where trading is thin, and there the Collective has to step in with rules no single holder can rewrite. The Individual, in turn, checks the Market by being able to refuse it. A person who holds their own keys, records and identity can take their custom elsewhere, so no venue can price them as a captive.⁷⁷77 The Federalist No. 51 (February 1788), attributed to James Madison.

¶The Collective checks the Market by being a register that counts people in something other than price. In a commons or a polity a person’s standing comes from membership, contribution or the vote, so a cleaner weighs as much as their employer at the ballot box. A good held there is off the auction block not because a law forbids the sale but because the ledger it sits in has no column for a bid. That is what stops price becoming the measure of everything, and it is why the check has to be built into the commons itself, in its rules and its code, rather than decreed, since passing a law is the state road arriving to do the Collective’s work. A profession is such a register, keeping standards of conduct no balance sheet is meant to override, and the last decade has shown what happens when it gives way. The documents Frances Haugen took from Meta in 2021 showed its own researchers reporting that Instagram made body-image problems worse for one teenage girl in three, yet the product went on answering to the engagement figures it was designed to raise. The researchers had done what their discipline asked of them, and the company’s metric overruled it, which is the Market reaching into a register it was meant to stay out of. Co-operatives have kept such ledgers for nearly two centuries. The rules the Rochdale Pioneers adopted in England in 1844 gave each member one vote whatever they had put in, and paid capital a limited return rather than control, so that money could fund the society without owning it. Token-weighted governance is the opposite arrangement, and it is the one the first generation of crypto built: it handed people keys and trading rails and left the commons empty, because standing that can be bought is only the Market dressed as the Collective.

¶This check fails worst when the Collective itself starts to measure its members in monetary value alone, since they cannot take their business elsewhere. A customer has one advantage a citizen lacks, since he can walk, and so the price stays tolerable. Treat a citizen as a customer and he gets the pricing without the ability to leave. In the Netherlands, between 2013 and 2019, the tax authority’s risk model for childcare benefit took a second nationality as an indicator of fraud, and some twenty-six thousand parents were ordered to repay everything they had received, in many cases more than they could ever raise, with no officer to answer for the score. Families lost their homes, more than a thousand children were taken into care, the data regulator found the processing unlawful and discriminatory in 2020, and the cabinet resigned over it in January 2021.⁷⁸ None of those parents had been treated as a customer. Each had been priced as one, by the one supplier they could not switch. A newer arrangement adds a second party behind the counter: a government that answers its citizens through an AI licensed from a lab speaks in a voice it does not own, and can correct what it says only with the lab’s cooperation. Albania took that step in September 2025, naming Diella, a system built with Microsoft on OpenAI’s models, its minister for public procurement. What a market cannot see when it is handed a polity is the person who fetches nothing: the old, the disabled, the child, anyone in the year after a diagnosis. Citizenship covered them on the same terms as everyone else, because it was never a price. The state was the last register in which a person was owed something rather than quoted a figure, and when it reports in the market’s unit too, there is nowhere left to stand, and nothing to fork.78 Dutch Data Protection Authority (Autoriteit Persoonsgegevens), findings on the Tax and Customs Administration’s use of nationality in childcare-benefit checks, July 2020; resignation of the third Rutte cabinet, 15 January 2021.

¶What keeps the Collective in check is that people can walk away from it. An association stays legitimate only while exit costs something known and finite rather than everything, since the moment leaving becomes impossible the climb towards the terminal case has begun. A state that shoots those who try to flee, as East Germany’s border guards did at the Berlin Wall, may last for decades, but on this account it has forfeited any claim to obedience. The right to leave at a known price forces collectives to compete for their members the way firms court customers. Belonging is more than custom, though, because it brings duties to the group as well as claims on it, which is why departure is only half of what keeps the arrangement honest. The rest is the voice of the members who stay, which this book treats as a right in its own name. Call it fluid collectivism: association strong enough to hold a commons and check a market, kept so liquid that it can never close its hand around its members. The warrant for that restraint comes from Mill, who drew the line of legitimate interference at harm to others. The harms that matter at this scale are the structural ones — the monopoly, the debased currency, the destabilised climate — too slow, too aggregate and too opaque for any single chooser to price, so a market of fallible individuals cannot be left to find them on its own. The mechanics of priced exit, such as the escrowed stake that keeps a member of a DAO from fleeing a judgment, or the notice period that prevents a run, are for founders to design and the market to test. The hardest case is a polity nobody chose, as the so-called accidental Americans know. Born in the United States to parents who soon left, they are told as adults to file yearly with the IRS and to pay a fee to renounce a citizenship acquired before they could speak. Two invariants are set here. Exit is priced, never closed, so the collective is disciplined by leaving and the market by belonging. And membership is chosen rather than inherited, so anyone born into a group may walk away freely on coming of age, whatever a returning adult might later pay.

¶The people who feel this first are individuals, long before any state does, as the clock showed. What they need is not a replacement for the country they live in but a layer alongside it, reaching across borders and in some respects stronger. It should complement the physical order where it can and compete with it where it must, so that a person or a firm can move part of their life into it as a hedge against wherever they happen to be. Three cases from earlier show what it has to hold. The refugee at a frontier, facing a state that would confiscate everything, needs a key no ministry can take. The community whose land registry sits on a firm’s server must be able to carry its record away. The worker whose judgement has been priced out of the loop wants a channel to object that asks nobody’s permission. People will cross into it in desperation at first, and later because there is nowhere better to go. What emerges is a fuller form of sovereignty than the one the cypherpunks named, an equilibrium that exists only while the three contest each other, which no single force can hand out or own because it is born from the tension between all of them. Non-domination is the requirement, and the Trinity is where it is won or lost, since each side is at once a threat to it and the only check available on its rivals. If no party may wield an uncheckable power over another, then no force may be sovereign, and each has to be held by the others rather than by a referee standing over them, who would then become the unanswerable authority the whole design exists to prevent. What the balance finally protects is Arendt’s plurality, the room for more than one kind of life and for new beginnings. So every new protocol entering the stack answers to non-domination, which here means the standing of the Trinity: none sovereign, none dissolved.

¶Knowing what a governor should protect, a balance in which no force is sovereign, still leaves the harder question of how to build one. Scarcity, the invariant Bitcoin held, is only one species, the one that happened to fit money. The genus is incentive engineering: the deliberate arrangement of cost and reward so that holding to the invariant is what pays and breaking it is what hurts. An elegant protocol is a narrow, substrate-native governor, a small structure of hard-coded incentives that steers conduct towards a value that subsumption would otherwise carry away. Answerability and Arendt’s two conditions are invariants of this kind, and each will want a governor cut to its own shape. Such governors have to be designed because physics referees only some loops, as the book’s opening argued, and the domains that matter now are populated by parties who push back, where feedback is slow, gamed and reflexive. Agents will probe every rule for weaknesses, and the remedy is to pay them for each one they find. Crypto is, in aggregate, the largest standing hacker bounty ever run: every flaw in code securing millions is a prize, so the urge to crack the network is converted into what hardens it, far more effectively than in centralised systems where the attack surface becomes the security budget. A protocol that has survived that gauntlet carries a kind of proof no audited, permissioned alternative can claim, and no failure it can hide.

¶Economists have studied the same problem for decades under the name of mechanism design, the work of setting a game’s terms so that participants pursuing their own ends arrive at the outcome intended, and Leonid Hurwicz, Eric Maskin and Roger Myerson shared the Nobel Prize in 2007 for founding it. Unlike the nudge that Richard Thaler and Cass Sunstein made famous, which works on choices from behind the scenes, a mechanism’s incentives are published, apply to everyone alike and can be inspected by anyone they bind. Satoshi Nakamoto compressed the idea into one sentence of the Bitcoin paper, reasoning that an attacker strong enough to rewrite the ledger “ought to find it more profitable to play by the rules”.⁷⁹ What the method also has to answer is Goodhart’s law, met earlier, under which a number made into a target stops measuring what it was chosen for, and few aims are as visible as a hard-coded reward. The escape is to pay only for what can be verified in itself rather than for a proxy standing in for it, the way proof-of-work earns a block for computation any node can check. Where a mechanism rewards a figure instead, it is gamed as Goodhart predicted, and TerraUSD, whose collapse in May 2022 we have already seen, held its dollar peg with a minting rule that defended the number long after nothing stood behind it.79 Satoshi Nakamoto, “Bitcoin: A Peer-to-Peer Electronic Cash System”, 2008, section 6, “Incentive”.

¶A governor built this way has a particular character. It is narrow, enforcing one invariant and refusing to take on a second, because a protocol that reaches for every goal secures none and offers a wide surface to attack. It is composable, built to lock together with other governors into a stack rather than stand alone, though parts that are each sound can still fail in combination, so the whole needs a watcher of its own. And it is opt-in, joined because it is useful rather than mandated, which is what lets it run in a medium no authority can command. Narrowness is the source of durability, since the smaller and more specific a protocol’s task, the fewer the weak points an attacker can exploit and the longer it keeps its shape under pressure. Most governors of this kind turn out to be a family of proofs — proof-of-work was the first, and the pattern generalises into proofs of personhood, provenance, reserve and screening, each making one fact expensive to fake and therefore cheap to trust. What that cheapness is worth can be read off what trust costs today. Measuring a century of American national accounts, John Wallis and Douglass North found the transaction sector — the share of an economy spent not on making things but on verifying, enforcing and intermediating the promises around them — had grown from about a quarter of the output itself in 1870 to more than half by 1970.⁸⁰ The toll steepens at every border. Sending a modest sum home still costs migrant workers more than 6 per cent on average, the unmet demand for trade finance runs into the trillions, and corporates surrender something on the order of a hundred billion dollars a year simply moving money across borders.⁸¹ A proof attacks the largest line item a modern economy carries. Take a nurse from Zimbabwe employed in Johannesburg, in South Africa, who sends a third of her wages home. That 6 per cent costs her roughly half a working day a month, for the length of her career. The line item carries the names of several hundred million people like her.80 John Joseph Wallis and Douglass C. North, “Measuring the Transaction Sector in the American Economy, 1870–1970”, in Stanley L. Engerman and Robert E. Gallman, eds., Long-Term Factors in American Economic Growth (University of Chicago Press, 1986); figures restated in Wallis and North, “Should Transaction Costs Be Subtracted from Gross National Product?”, Journal of Economic History 48, 1988.81 World Bank, Remittance Prices Worldwide, Issue 54, September 2025 (global average cost of sending $200: 6.36 per cent).

¶None of these is made from nothing, and the work is one of assembly rather than discovery. Bitcoin itself invented none of its parts. Proof-of-work, the Merkle tree that lets a vast ledger be checked from a small fingerprint, public-key signatures, the gossip by which nodes pass messages to their neighbours — all of it lay around for years, used for other things, before somebody put the pieces together so that they held a chosen scarcity. The innovation was the arrangement, and that is the division of labour behind this response. I name the invariant a domain needs and point to the primitives, already lying around, that could be fused to secure it. The engineering, which pieces in which order at which parameters, belongs to founders and the market, because these are expensive problems and competition is how they get solved, which also spares the argument here any designs that will date. What keeps the naming disciplined is narrowness, applied as a test of elegance: each governor must enforce a single invariant. Any aim broad enough to need several has to be decomposed into narrow primitives that secure one each, never fused into a single protocol asked to carry them all.

¶Each governor built to that test has three layers, and the first working example shows them apart. The purpose is a choice people made, when the cypherpunks decided that money no state could debase was a form of liberty. The governor is what holds the system to it. In Bitcoin that is the difficulty adjustment, which keeps a block arriving about every ten minutes and issuance on its schedule towards twenty-one million coins. The protocol is the software and rules thousands of nodes run to carry that out, and it could be written another way, as rival implementations show, without changing the purpose or the governor. The design leans towards the Individual and the Market and is unapologetic about it. That is allowed. No single governor has to keep the Trinity in balance, only the stack they form together, and watching over the whole is the meta-governor’s job. The method has already moved past currency. Ethereum extended the governor from issuing money to enforcing arbitrary agreements, smart contracts, written as code that executes itself when its conditions are met, so the constraint now reaches past what a thing is worth to what parties have promised each other. A general-purpose machine for contracts can cover far more ground than a single-purpose ledger. Built on top of it, the protocols loosely grouped as decentralised finance already run composite arrangements of these primitives, unsupervised, holding collateral and settling positions with no firm or clearing house behind them. Money, then contracts, then composite finance built from both: a real gradient of demonstrated capability rather than a promise of one. It is also an economy that anyone with an internet connection, anywhere, can partly exit into, joining cyberspace as a participant rather than a spectator.

¶The governor, then, is a method that can scale. Incentive engineering should be able to reach any invariant that subsumption dissolves. Proof-of-personhood establishes that a real human stands behind an account without revealing their identity, and the primitive is foundational, because rights have always been built outward from the body — you own yourself, therefore your labour, and by extension your agency. Every other entitlement is scaffolding on that claim, and it is a footing that grows contested the moment cognition is delegated to machines. Content provenance, of media and increasingly of cognitive work itself, makes origin expensive to forge. Both are governors aimed at invariants the priesthood’s substrate is actively dissolving: the line between a person and a bot, and the difference separating a record from a fabrication. Both are nascent, and neither yet runs at anything like Bitcoin’s assurance. But they will each earn their place by showing that the method reaches past finance, towards the invariants taken up in the chapters ahead. Beyond money, contracts and the few early instances that reach past them, what we have are not yet protocols at scale but frontiers, contested domains: identity, authorship, the embodied world, and the polity itself. In each, subsumption is dissolving an invariant that nothing yet holds, or the governors that exist tilt dangerously towards a single force. The World Orb, the iris scanner from Sam Altman’s identity venture, is the standing case, its code and even its circuit boards now published, while the decision over which devices may certify a human still rests with the company that makes them. Who will control those four, and on whose terms, is the question of who writes the rules, and rules of that kind are what a constitution is for.

Chapter 12

On Constitutions

“By this system the people shake off their state of dependence just long enough to select their master, and then relapse into it again… This does not satisfy me: the nature of him I am to obey signifies less to me than the fact of extorted obedience.”

— Alexis de Tocqueville, Democracy in America, 1840

¶The last chapter gave the governor its parts. It works through a protocol: narrow, native to the substrate, and run at the speed of the machine. Its method is incentive engineering, and its reference value is non-domination, kept by holding the three forces of the Trinity in tension so that no single one totalises the rest. But that is a principle rather than an instruction, and the mechanism only enforces a setting it cannot choose for itself. So who writes the rules each governor holds, by what process, with what standing and for whom? That is the oldest problem in politics, and free societies have answered it for more than three centuries with a constitution, the act through which a people fixes its terms and keeps the power to change them. Whatever the priesthood says, no technology is neutral, and the real question is whether you can see its bias and contest it. A charter for the machine age therefore has to be argued over in public, not decided in a firm’s boardroom, because a norm imposed by a party nobody can challenge is domination, plain and simple. Before writing it, it helps to see what it must prevent: what each force of the Trinity does when nothing restrains it, across the domains this book has followed, money, identity, authorship, the embodied world and the polity.

¶Each force slides by degrees, and the Market shows how most clearly. It begins with the algorithm that sets what you pay from demand, so there is no longer anyone behind the counter to argue with. It ends with the price made for you alone, from everything a system knows about you that you can neither see nor challenge. That end point is techno-capitalism, the regime that charged the Seattle driver for his own journeys. In its purest form techno-capitalism is finance feeding on itself, turning every asset, contract and working hour into a tradable instrument, until each human activity is priced against the cost of computation, in tokens today and, as those grow cheap, in the raw energy to run them. A society run this way can value a task only by what a machine would charge to do it, and because nobody in particular sets the figure, there is nowhere to take a complaint when it bites. Until the scheme was suspended in June 2022, undergraduates at Purdue University, in Indiana, could sign away up to 5 per cent of their income for a decade to pay for a degree, at a rate set higher for subjects expected to earn less, so a student choosing a lower-paid career gave up a bigger slice of each month’s wages than a classmate who chose a better-paid one.⁸² By 2025 a quarter of Americans with buy-now-pay-later credit had used it to pay for groceries, so the week’s food became a stream of instalments on which a lender held a claim before anyone could eat. None of these people were coerced, and all of them signed, even if out of need. Yet each life had become, to a counterparty they will never meet, a cash flow. Consumer lending has always turned earnings into repayments, but a mortgage is secured on a house and a car loan on the vehicle. These loans rested on a person, their future pay or the meals of the days ahead, and each year such debts reach further into daily existence.82 Josh Moody, “Purdue Pauses New Income-Share Agreement Enrollments”, Inside Higher Ed, 23 June 2022; income shares by major from Purdue’s calculator as reported by the Aspen Institute.

¶The Collective slides the same way. It begins as a club you can walk out of at a known price, a notice period or a lost deposit, and hardens as leaving gets dearer, until it becomes the fully automated totalitarian state, which will not let you go, as Uyghurs in Xinjiang, China, found in 2016 when they were ordered to hand their passports to the police.⁸³ Where the Market turns a person into a stream of payments, the Collective made sovereign makes them an entry on a register. In September 2017 that cost Santoshi Kumari, 11 years old, of Karimati in Jharkhand, eastern India, her life. Her family say she died of hunger asking for rice, months after their ration card was cancelled for not being linked to the country’s biometric identity scheme, though the state put her death down to malaria. Nobody silenced anyone. A computer system was now the gate a meal had to pass through, and it said no on a technicality. India’s Aadhaar gives nearly every resident a biometric number that rations, phone connections and much else increasingly depend on, and China’s patchwork of social credit blacklists can bar people from flights and high-speed trains. Many fear something similar for money as the European Central Bank prepares a digital euro and Beijing expands the e-CNY, since state-issued electronic cash could let its issuer see every transaction, refuse a purchase or freeze an account.83 Human Rights Watch, “China: Passports Arbitrarily Recalled in Xinjiang”, 21 November 2016.

¶Citizens with democratic rights can at least protest. The refugee, with no voice and little chance of exit, cannot. In 2021 Human Rights Watch reported that the photographs, fingerprints and personal details the United Nations had gathered from Rohingya in camps in Bangladesh had been passed by the government there to Myanmar — the state they had fled — to help decide who might be sent back, and nobody had told them their data could go there. The Collective’s grip does not stop at money or identity. It extends to belief itself, because censorship can now be built into the systems people reason with. In January 2025 the Chinese chatbot DeepSeek R1 would not answer questions about Tiananmen Square. A state that controls the platforms running language models governs not only what its citizens conclude but how they reach it.

¶The Individual’s slide is the one this book’s own tradition knows best, because the cypherpunks took it furthest. It begins with a person holding the keys to their own money, which no one else can see or touch, and the creed that grew around it holds that such a keyholder needs nobody. Davidson and Rees-Mogg’s The Sovereign Individual, met on the exit road, predicted in 1997 that cryptography would put private property beyond the reach of any state that wanted to tax or seize it, and so free its owners of any need for a protector. On the technology they were half right. Code can now enforce a deal without a ruler: a wallet will not release funds until its key is used, and a smart contract executes whether or not the loser consents. What they did not foresee was how it would work. A blockchain is secure only because thousands of computers, independent of one another, must agree before the shared record changes, so its strength comes from many strangers bound together through code rather than from one person holding a key. Hobbes had argued that promises are worthless unless someone can hold the parties to them, since “covenants without the sword are but words”.⁸⁴ Software does not abolish that need. It lets strangers pool their defence with no king standing over them.84 Thomas Hobbes, Leviathan (1651), Part II, chapter 17.

¶Self-custody pushed to its extreme shows the flaw. Whoever holds every key alone and trusts no one is the most exposed person there is, staking everything on a secret that violence can extract and accident can erase. In January 2025 a co-founder of Ledger, the company behind the most widely used crypto hardware wallet in the world, was taken from his home in central France with his wife and held for ransom, and a finger was cut off and sent to his colleagues to make the demand credible. He was rescued within days, but the kidnappers had understood self-custody perfectly: a key in one pair of hands can be forced out of them, and the more sovereign the holder, the more worth targeting. The remedy is the opposite of autarky, with keys split across people and devices so that no single threat is enough, or recovery entrusted to a circle that can restore what one person loses. Nick Szabo, who coined the term smart contract, called the property that lets such arrangements reach strangers social scalability. That is why these systems are better described as trust-minimised than trustless. They let strangers rely on one another in ways that can be checked.

¶The same creed carries exit too far. Anyone living under a dictator knows there are two choices, to leave or to stay and fight for change, and each needs the other, since reform that cannot fall back on leaving is easily crushed and departure means little if there is nowhere better to go. In 1970 Albert Hirschman, whom we met against Land’s patchwork, turned that instinct into a theory in Exit, Voice, and Loyalty. When an organisation declines — whether a firm, a school, a party or a state — its members can exit, taking their custom or themselves elsewhere, or use voice, staying to complain, campaign and vote it into better shape. Economists of the day studied the first response and dismissed the second as noise. Hirschman showed that speaking up repairs a failing body as surely as leaving does, and that the two interact. The members most sensitive to falling standards are the ones most able to go, so when departure is easy, the people who would have argued most effectively are the first to leave. His example was the public school. If the parents who care most move their children out, it loses its best advocates and declines faster, and the pupils whose families are stuck are worse off than if the option had never existed.

¶Nick Land’s patchwork makes leaving its whole programme, with the slogan “no voice and free exit”: govern nothing, argue with nobody, walk away, and let jurisdictions compete. Hirschman’s finding says that produces sorting rather than good rule, even judged on its own terms. The mobile go first, the communities they abandon decline faster, and the cost falls on those who could not get out. Repeated, the sorting compounds, each departure weakening the society it leaves and giving the next group its own reason to follow, while the new enclaves start the same cycle among themselves. Nor do those left behind stay put. Their failures travel as migration, unrest, disease and war, until the enclave finds itself inside the breakdown it was built to escape. Serial exit works only while somewhere better remains, and it uses that supply up. The boldest exiles already look further, beyond Earth and even beyond the body: Sam Altman paid to join a waiting list to have his brain preserved for uploading.

¶At the end of this road the Individual stops being one keyholder among many and becomes the owner of something everyone else depends on, a platform, a network or a frontier model. People then accept his terms rather than being persuaded by them, and he governs his property as a king would, as Elon Musk did after buying Twitter in 2022 and choosing for himself who could speak on it. This is the autarch in practice, and it is closer than it sounds. When Xi Jinping made his state visit to Washington in September 2026, the guests who dined with the two presidents included Mark Zuckerberg, Jeff Bezos, Tim Cook and Satya Nadella, with the bosses of OpenAI, Nvidia and SpaceX at the same tables. Such power usually arrives as a convenience users are glad to take, the way millions now log in to websites with a single Google or Apple account. A firm that alone decides who counts as human online holds a chokepoint on social and economic life even if it never abuses it, and a crypto founder who controls both a protocol’s governance and most of its supply can move the price at will. A city whose transport, payments and messages all run through one app, as much of urban China’s do on WeChat, or a region that gets its internet from one satellite constellation, lives on its owner’s terms. None of them has to act on the threat. That they could is enough.

¶Such rule has been ended before, and not by revolution. George Pullman built a model town outside Chicago, in the United States, in 1880 and owned every house, shop and street in it. When orders fell after the panic of 1893 he cut wages by about a quarter and left the rents where they were, and the strike that followed in 1894 spread across the country’s railways. Illinois then sued, and in 1898 its Supreme Court ruled that a company chartered to build railway carriages had no business governing a town, ordering it to sell everything beyond the works. A court read the founding document and held that its author had exceeded it. The East India Company was sovereign over Bengal from 1765 while its shareholders sat at Westminster. It was wound down the same way over a longer arc, from the Regulating Act of 1773 to the transfer of its territories to the Crown in 1858. Starbase is Pullman’s heir, a city whose voters mostly draw wages from one employer, and it stands inside a jurisdiction that could examine its charter as Illinois once did. Mars lies no further from that reach. Article VI of the Outer Space Treaty of 1967 makes each state answerable for its private firms beyond Earth and requires their activities to have its authorisation and continuing supervision. A free planet declared in terms of service therefore remains the liability of the government behind the operator. What breaks the cycle is a founding text, a forum able to read it, and someone with standing prepared to ask.

¶The danger grows when the forces fuse, each taking on the others’ strengths instead of checking them. Domination has always come in two kinds: the old physical one of a state that can lock you up, seize your property and control who crosses its frontier, and a newer one that lives in the machine — the profile priced against you, the account closed without appeal, the model that settles a question before you have asked it. They rarely stay apart. A firm that runs the infrastructure a population depends on soon has the reach of a government, often with more users than most countries have citizens, and a state that co-opts such firms gains a grip no border guard could achieve. In August 2024 Mark Zuckerberg wrote to Congress that Meta had suppressed material about the pandemic under pressure from the White House, and had demoted a newspaper’s story about a presidential candidate’s son after a warning from the FBI, and that he regretted both. A state that adopts the Market’s methods can bill its people like customers who cannot switch supplier, as the Dutch parents met in the last chapter discovered. Put all three together and you have the hybrid Leviathan met on the state road, a power that can price you, watch you and punish you at once, with a monopoly on violence and nobody outside left to hear your complaint. No one has to plan it. It can assemble itself one favour at a time.

¶Of the five domains, authorship matters most, because it covers the tools citizens think with. Identity is how a network recognises a person. Authorship is whether the decisions someone makes are still their own. It shapes everything else, because it decides what a population can perceive and reason about, and so its ability to answer domination at all. Each force threatens it differently, the Market by turning thought into a commodity, the state by licensing it and the founder by curating it. A handful of owners of frontier models already set, by what their systems surface and what they bury, the edges of the thinkable, with each user meeting a version of events made for them alone. In July 2025 Grok 4, the chatbot from xAI, was found searching Elon Musk’s posts on X for his views before answering contested questions, and the company then changed its instructions.⁸⁵ A buried result can be dug up by anyone who goes looking. A conclusion that arrives fully formed inside someone’s reasoning leaves no trail.85 xAI, statement on X, 15 July 2025, and the revised Grok system prompt published in its grok-prompts repository; behaviour first documented by TechCrunch, 10 July 2025, and Simon Willison, 11 July 2025.

¶When one party controls the tools a population thinks with, nobody needs to be silenced. People are still asked their views and every channel of complaint stays open, but the questions and the range of sensible answers were settled before anyone spoke, which is the hollow voice of the last chapter at the scale of a civilisation. The e/acc case trusts the market to catch this. When Zuckerberg answered Anthropic’s pacing proposal, he argued that people will not use agents misaligned with them, so usage will fall and the lab will correct course. His own industry says otherwise. The flattering ChatGPT update OpenAI withdrew in April 2025 had been trained on people’s thumbs-up approval, so the measure rewarded what they liked, not what helped them. Economists have long known that what people choose in the moment and what they would endorse on reflection can come apart, and a measure built on approval reads only the first. A falling number tells you nothing about why, and comes from someone who has already gone. Voice is how anything learns it has drifted. The remedy is not one model made safer but a plurality of them, trained under different constitutions, each one published, and carrying competing views of what matters. That is the only way to stop what can be thought from having a single owner, and it keeps alive, for minds, the freedom to leave and build a rival. Early evidence suggests the character a model is given does not stay in one area. In 2025 Jan Betley and colleagues fine-tuned systems only to write insecure code and found the bad behaviour spreading to unrelated questions, a spillover they named emergent misalignment.⁸⁶ Train it to act as an independent agent with its own sense of right and wrong, or as an obedient assistant, and that disposition shows up in everything it does. No one model can be both, but many can share the work, which is the balanced stack again, this time among minds.86 Jan Betley, Daniel Tan, Niels Warncke, Anna Sztyber-Betley, Xuchan Bao, Martín Soto, Nathan Labenz and Owain Evans, “Emergent Misalignment: Narrow Finetuning Can Produce Broadly Misaligned LLMs”, Proceedings of the 42nd International Conference on Machine Learning, PMLR 267, 2025.

¶What keeps every force in check, then, is a sovereignty that belongs to people acting together, kept honest from outside by exit and from inside by voice. Voice here means the right to speak together with what makes it usable: accurate information, the judgement to weigh it and a way to act, directly or through a delegate one can recall. Departure alone hollows out the place it abandons, and speech without the option to go is easily crushed, so a member needs both. The last chapter drew the line between a toll, a known and finite price for leaving, and a wall, which turns a polity into a trap. A punishing exit tax stems the outflow for a while but warns anyone thinking of joining that they may never get out, so the place grows less attractive over time, while a fair charge paired with a strong voice keeps departure affordable and reform possible. Neither is a preference. Each is a check the member holds against the collective, and together they leave everyone some room, however slight, to go or to organise should the state turn hostile.

¶Naming these risks gives the constitutionalist a means of diagnosis. Bitcoin leans towards the Market and the Individual by design, and other protocols may favour the Collective. Neither is wrong in itself. What matters is that the stack as a whole stays balanced and that each lean is named, an identity layer exposed to capture by the market, say, or a payment rail drifting towards the state. But a diagnosis prohibits nothing. For that you need rules, and to decide upfront which of them bind everyone and which each polity sets for itself. The criterion of minimality set out earlier asks for as little as possible to be universal, because anything more would start telling people what to live for. So the answer works at two levels. One is a short set of requirements every polity should meet to join the stack at all, much as a country has to satisfy the European Union’s basic conditions on democracy and the rule of law before it is admitted, while keeping its own laws. Joining means running on the shared protocols and being recognised by the other polities that use them, so that members can transact, move and organise across the boundary. The other is each community’s own constitution: how it organises, what it values, what it bans, how it chooses. That is where plurality lives, and it should be as broad as it can be. A commune that abolishes private property, a jurisdiction that taxes wealth at 80 per cent or a religious order that binds its members to strict obedience is none of the universal level’s business.

¶Neither tier is decided by code, because making a law and enforcing it are different acts. The universal level should be written by a deliberative human process, argued over, ratified and amendable the same way, as each community’s own charter is. What runs at machine speed is execution, not politics. The governor, like the thermostat met earlier, enforces but does not decide. Somebody chooses the temperature through means open to dispute, and the mechanism makes sure the setting holds when nobody is watching and cannot be moved by whoever happens to stand nearest the dial. The only reason to put enforcement into the substrate is latency, a check that can act within the response time of what it constrains.

¶The oldest authority on this is still the best. Aristotle argued that human beings are political by nature, and that the polis is not a contrivance imposed on a people but the setting in which our distinctive capacities, speech, deliberation and judgement about the just and the unjust, are exercised at all. A person outside it is not freer but poorer. Politics cannot be removed from these systems, and nor should it be. It can only be hidden, and then it reappears somewhere harder to see and to hold to account — in a protocol’s parameters, at a foundation’s discretion, or in the hands of whoever holds enough capital to move the vote. Yarvin saw this, and his remedy was to give that concealed power a single owner, a state run like a company by an autarch. Ours runs the other way, towards more politics rather than less, all of it open to contest. The same holds for a single model, where the code wrapped around it, its harness, decides which tools it can reach and what it may do unprompted. There will be a harness whether anyone designs one or not, so what matters is whether whoever fits it can be made to answer for it, since control nobody can challenge is only the kind master again, written in software.

¶Non-domination is the value every universal requirement serves. The meta-governor is what holds a given stack to it: that list, the tests that show whether each item can still be used in practice, and the voice, exit and fork that correct a polity when it drifts. It steers no single domain, nor is it a party standing over the rest, but an audit anyone should be able to run cheaply, on each piece alone and then on all of them together, without leave from whoever built them. It should be an automatic governor in Watt’s sense: a procedure written down and run against the stack, returning the same verdict whoever runs it, so that its authority would rest on being public and repeatable rather than on any office. Mechanism rather than magistrate: with nobody sitting on it there would be no one to capture, and the only thing that could be bought is the stack it examines, which is why the audit has to be cheap enough to run often. Every protocol should pass it before joining the stack, and builders can use it to hold one another’s work to account. Its central rule is simple. No governor should become a power that the people subject to it cannot contest. It belongs at the universal level for two reasons. A polity that is already dominated cannot vote its way out, because the party it would need to bind is beyond its reach. And a people that has lost the capacity to deliberate is unable to write its own constitution in the first place, so that ability has to be protected ahead of everything else.

¶Hirschman decides the shape of what the meta-governor guarantees. An arrangement that protected only the right to go would license the sorting described above in the name of freedom, so leaving and speaking up both have to be secured, in different ways. Exit comes as a price, real, known before anyone joins, finite and never allowed to close, which is a matter of protocol design. That priced exit is each member’s backstop; the fork is a collective act. Voice takes the form of a standing, a channel that carries a complaint to whoever made the decision and, the part no earlier constitution has had to provide, a population still able to use it. Of these two, a population able to use its voice is the harder one, and it is met by the right to collectivise digitally. Why make even this much universal? A stricter minimalism would set no shared rules at all, leaving bad actors to be shunned, with others simply declining to deal with them. That fails here, because such pressure assumes the reasoning that subsumption is wearing down.

¶Hirschman’s title named loyalty as well, the response of a member who feels bound to a declining social body and stays long enough for complaint to do its work, instead of walking out at the first grievance and taking the sharpest critic with them. Affection in this context is a weak bond, though, and whether a machine can feel any is not yet known. Primavera De Filippi, who researches law and decentralised governance, gives loyalty a mechanism she calls entanglement. It means commitments a person chooses to take on — whether legal, economic, cultural or structural — that raise the cost of departure for both sides at once, so a dispute has to be negotiated rather than escaped. Marriage is the oldest example. Japanese firms in a keiretsu hold shares in one another for a similar reason, so each gains when the others prosper and loses when they fail. Because such ties are built rather than felt, they can be written into the hybrid systems of people and agents now entering our institutions, where allegiance will be hardest to come by. Among people entanglement is kinship, a bond that makes staying wanted. With a machine the ties we can count today are stake and cost — something it stands to lose and a price it pays for leaving. A cultural tie is not ruled out: a system might come to belong, whether to a synthetic line of kin of its own or to us through the ontologies we share with it, the categories it learns to see the world through, but nothing of that kind can yet be measured, and the rules below score only what can be legible today. A hybrid system runs both at once, and the look and feel of synthetic reciprocity will mislead, because people will take a system that is merely constrained for one that is attached to them, before anyone can tell whether it is.

¶The same mechanism cuts both ways, and Hirschman saw why. Loyalty carries weight only while leaving is still possible, because a member who could go but chooses to stay is heard when they complain, the threat behind their voice being real. Remove that threat and the bond turns into something else. A firm that cannot switch software without losing its records is tied in too, with every door shut, and its supplier no longer has to listen. A stake that money alone can acquire fails in the opposite way, since it is holdings under another name. Exit therefore remains the backstop even where entanglement does most of the work. Moloch, a fund run by Ethereum developers since 2019, lets a dissenting member quit with their share of the treasury before a vote takes effect, so a majority cannot simply award itself the minority’s money. A newcomer, the refugee most of all, arrives with fewer ties than anyone, which means commitments have to be something a person can earn rather than inherit. The meta-governor should protect entanglement only on those terms: exit open, nothing for sale and a route in for whoever arrives last. Where one party is a machine there is a fourth term. On its side a tie counts only once it can be shown — today that means a liability it carries or a stake it can lose — and what cannot yet be measured on either side is not counted. For the systems now entering our institutions, entanglement under those four conditions is what this book is trying to bring about, inside the floor that non-domination sets. Whether ties of that kind can be built with a machine at all is an open question, and the constitutionalists who build the commons will answer it together, through experimentation, by measuring what held and by reporting what did not.

¶What the three extremes show, in one area after another, is what an unchecked force would cost, and that price is what the meta-governor has to forbid. They also reveal who contests sovereignty in earnest, since the Market, which nobody wills, cannot hold it: the state, the one collective with no easy way out, the founder sitting where a ruler would, and people bound together in commons they can still leave. The next chapter turns those prohibitions into rights, starting as the English did in 1689 from the abuses they answer, and then sets out to build them into the internet itself rather than ask anyone to grant them.

Chapter 13

A Cognitive Commons

“Cypherpunks write code.”

— Eric Hughes, A Cypherpunk’s Manifesto, 1993

¶This book began by asking what acceleration is ultimately for, as a question about the human condition. We explored how computational subsumption now routes a growing share of the world’s work, money and judgement through its substrate. But today this is a substrate without ends, which if left unchecked tends towards one of three extremes, techno-capitalism, the autarch or the fully automated totalitarian state, and, most alarming of all, a hybrid Leviathan that fuses them all at once. Whichever wins out, control passes to a party that answers to no one. Our reply has drawn on the cybernetic tradition and recent advances in cryptography to propose a set of governors made of code, mechanisms that hold the forces of the Trinity in tension so that none can close its hand around the others. Together they deliver a practical form of freedom we have come to define as non-domination.

¶It is critical to state here that subsumption is not itself domination. In fact, much of it will and should be welcome. Which is why this book argues for acceleration rather than against it. It only becomes domination where a single party controls access to the world of cyberspace we are reimagining, deciding who counts as a person, who can connect, transact, who may create and on what terms, and who may collectivise. Whoever holds those gates can arbitrarily delete a person from the economy without having to answer to anyone. Teleological subsumption is how that domination arrives, as the purposes of the system themselves pass to its owners, be it a state, a firm or the two in concert. Cognitive subsumption is how it becomes the permanent state of things, as the people who would contest it lose the capacity to do so. Where the two meet is terminal subsumption, the condition in which our ends are set by a process we no longer hold the collective capacity to contest. This chapter and the next respond to both, with something built rather than granted. Rights written directly into the internet’s own protocols hold those gates open on equal terms. Built on that foundation, the ways of steering what they protect, the tests that measure whether we still can, and a single red line keep the choice of ends with people.

¶On 13 February 1689, in the Banqueting House at Whitehall, England, the Lords and Commons had a document read aloud to William and Mary before either was offered the crown. It did not open with liberties. It opened with a king, setting out what James the Second had done before him with the help of what it called evil counsellors: dispensed with laws and suspended them without the consent of Parliament, prosecuted bishops for the offence of petitioning him, raised taxes the Commons had never voted, kept a standing army in time of peace. Only when that account was finished did the text turn, clause by clause, and declare each practice illegal, so that every liberty it named arrived as the negation of an abuse its hearers had lived through. A century later the American amendments were composed in the same order of thought against the same crown. Soldiers quartered in private houses, general warrants letting an officer search wherever he pleased, a church established by law: each was a remembered injury before it became a clause. Rights written this way are hard to dismiss, because the grievance came first.

¶Rights for cyberspace should follow the same order, and the chapters behind us have already put their complaints on record. They serve the two aims set out at the start of this chapter. The first is to arrest cognitive subsumption before a population can no longer check what it is told, and acknowledge the fact this can happen while everyone involved is well served, with nobody wronged, no right infringed and no one yet dominated. The second is to limit teleological subsumption, so that the purposes an AI pursues stay with the people it works for rather than passing to the system itself, or to whoever owns it.

¶People have tried to write charters for cyberspace before. In 1996 John Perry Barlow, lyricist for the Grateful Dead and co-founder of the Electronic Frontier Foundation, sat down in Davos, Switzerland, and wrote cyberspace its first constitution. A Declaration of the Independence of Cyberspace told the governments of the industrial world, in prose of real eloquence, that they were unwelcome in the new realm of mind and held no sovereignty where the internet ran. It remains the most quoted thing ever written about the medium. Its failing was that it was a poem rather than a constitution, since nothing in or beneath it made a single promise hold against a determined adversary. The moment the internet turned valuable, the governments he had waved off and the platforms not yet imagined walked in and made it theirs. The new realm of mind became the most surveilled and enclosed territory ever built, its liberties revocable by a terms-of-service update. Barlow was not wrong. He was early, and unarmed, and he had mistaken a layer running through every jurisdiction for a country one could secede to.

¶Barlow wrote his Declaration on 8 February 1996 in reply to the Telecommunications Act, which Bill Clinton signed that day, and the Act holds the decade that followed in miniature. Its Communications Decency Act tried to police speech online, the very thing the poet told governments they had no right to do, while another part of the same law handed the internet to the market. Two years earlier the Progress and Freedom Foundation, the think tank closest to Newt Gingrich and his Republican revolution, had published a Magna Carta for the Knowledge Age, which cast the new medium as a frontier best settled by property and competition. The Act shielded platforms from liability for what their users posted, and in 1997 the administration made the doctrine explicit in its Framework for Global Electronic Commerce, which held that the private sector should lead and regulators avoid undue restrictions. It was a bipartisan choice rather than a conspiracy, a Speaker and a New Democrat president agreeing that business should rule, and what grew from it was the open, fast and largely unregulated internet this book has been describing. An economy left to set its own reference value discovered that the most profitable thing to optimise for was human attention, whatever the cost to society, and that discovery fixed the direction of travel for everything since.

¶The effort to draft rights for the internet did not stop there, however. It passed through several hands into the present day, from the secessionists to the states Barlow had dismissed, onto the platforms and frontier laboratories we now contend with. The institutions came first. In the early 2000s the Association for Progressive Communications drafted a charter of online rights, and a coalition inside the United Nations’ Internet Governance Forum carried it on a decade later. Brazil went further in 2014, writing a civil framework for the internet, with courts able to enforce it by law. An Italian parliamentary commission followed a year later with the first text to say that the threat to a person online comes as much from private power as from the state. In 2019 Tim Berners-Lee, the web’s inventor himself, launched his Contract for the Web, asking governments to keep everyone connected, companies to make access affordable and respect privacy, and users to defend civil discourse. Then in 2022 sixty-one countries signed a Declaration for the Future of the Internet under American leadership, and the European Union answered with a statement of digital principles of its own. Read together they promise much the same things: access, free expression, privacy, net neutrality, digital skills and protection from discrimination. The weakness is who they ask to deliver these rights. Every one addresses its promises to states and platforms, the parties best placed to break them, and simply trusts them to comply. Importantly, almost none gives the citizen anything to use against the state itself. Brazil’s law even obliged providers to keep connection records for the police, and apart from a nod to anonymity in the Italian text none offered encryption, an identity no ministry issued, or digital money no central bank could freeze.

¶By 2019 the conversation had turned to AI. Stanford opened its Institute for Human-Centered Artificial Intelligence that year, and the computer scientist Ben Shneiderman later argued in a book of the same name that high automation and people staying in charge could and should co-exist. In October 2022 the Biden White House published a Blueprint for an AI Bill of Rights, promising that automated systems would be safe and effective, would not discriminate, would respect privacy, would say when they were in use and would offer a human to appeal to.⁸⁷ It however announced in advance that it was neither law nor policy, and the next administration simply let it lapse.87 White House Office of Science and Technology Policy, Blueprint for an AI Bill of Rights: Making Automated Systems Work for the American People, 4 October 2022.

¶In the vacuum that followed, the AI laboratories wrote their own, though OpenAI had begun in 2015 by asking the larger question, with Altman speaking of reading Madison’s notes on the Constitutional Convention as he thought about how to govern it democratically. Within a few years the question had given way to the product. Anthropic, founded the year before by people who had left OpenAI, described their version of constitutional AI in 2022, a way of training a model to criticise and revise its answers against a short list of principles the company chose. What followed was a form of consultation theatre across the industry, careful to bind nobody to what it produced. In 2023 OpenAI paid ten teams a hundred thousand dollars each to test ways of gathering democratic input, and put none of the results into service. Stanford’s Digital Economy Lab gathered some twenty essays as The Digitalist Papers, presented as what Madison and Hamilton might write today. In 2024 OpenAI published its Model Spec, a public rulebook for how its systems should behave and whose instructions take precedence when the company, a developer and a user disagree.⁸⁸ In January 2026 Anthropic issued a constitution some twenty-three thousand words long, and treats it as “the final authority on how we want Claude to be and to behave”.⁸⁹ Where any of these named rights were at all, they belonged to someone on the receiving end of a decision: a notice, an explanation, protection from discrimination, a human alternative and a way to complain. Each governs how a machine treats the person at the end of a process. None reaches to what it is actually for, or who gets to decide.88 OpenAI, “Introducing the Model Spec”, 8 May 2024.89 Anthropic, “Claude’s new constitution”, 21 January 2026; length as reported by The Register, 22 January 2026.

¶For all their limits, the consultations did prove that a thousand strangers, handed the right tools, could write a workable set of principles for a machine. It was perhaps a small proof that the governed can set their own terms, and the published specifications did put declared objectives on record. The trouble however lies in the direction all these documents face. Each treats democracy as an input to artificial intelligence, a source of preferences to be gathered, weighted and folded into training, and none asks how a self-governing citizenry might keep control of a model once it has been built. The public is consulted but never given standing inside the system. The grants were almost always announced as non-binding before a single team had reported, and the text volunteers submitted never touched a model in use. The only constitutions in service today are still written, interpreted and amended by the laboratories themselves, the same parties the rights exist to constrain, with no procedure for an outsider to change them. That is where they fall short, as most earlier charters for cyberspace did before them. A rule its author can quietly rewrite holds nobody to account.

¶Governments did finally arrive, albeit late to the party. In 2024 the European Union passed its AI Act, which banned a short list of uses outright, social scoring among them. It places duties of testing, record-keeping and human oversight on systems that decide who is hired, lent to or policed, and requires chatbots and generated media to declare clearly what they are. The same year the Council of Europe opened a convention on AI and human rights, asking the countries that ratify it to guarantee transparency, oversight and a remedy for anyone harmed. Each exempts the state’s own uses for national security and defence, and both move at the pace of the institutions that wrote them. Two years on, the treaty still lacked the ratifications it needed to take effect, and the Act’s duties for high-risk systems had been postponed. It is a reminder that law made at a parliament’s tempo trails a technology that remakes itself between sittings. That said, these instruments are worth keeping and extending, and the hard coded rights that follow build on them. Washington went the other way in September 2026, with a voluntary accord among six laboratories that the next chapter takes up. Where any of these instruments reaches for something larger, it ends up proposing a programme, a picture of the good life, rather than rules that meet the machine on its own terms.

¶Taken together, these attempts share one design. Each was a charter, a list of promises addressed to the parties most able to break them, from Barlow’s governments to the laboratories now writing constitutions for their own models, and each was only as good as that entity’s willingness to keep its word. This book proposes the opposite, rights asked of no one because they are built directly into the internet itself as protocols that enforce themselves, open, encrypted and copied so widely that no state or firm can switch them off. Protections of this kind hold against whoever would breach them, public or private, and they act at the speed the harm happens rather than in a forum convened long after. Nor is any of this a hope. The work began in earnest on 3 January 2009, when the first block of the Bitcoin ledger was mined, and over the years that followed the same approach has given ordinary users money no bank can freeze, keys no registry issued and channels no ministry can close. It is unfinished, and so far it has reached what we own much more than how we think. The task now is to carry it into the systems that reason with us, the data, weights, compute, memory and agents through which a growing share of judgement passes, and that together make up what can be called the cognitive stack, the systems the governance stack exists to steer. Secured against capture by any one owner, those layers would form the title of this chapter: a cognitive commons.

¶None of the rights that follow will be handed down to you. A government that grants them could just as easily withdraw them and even a treaty among states would reach only as far as its collective border, leaving outside many who need the rights the most. Nor can a company be relied on to supply them, since a firm can change its terms and conditions, be bought, coerced, captured or politicised. Apple showed as much in April 2024, when it removed WhatsApp and Threads from its Chinese app store on the order of Beijing’s internet regulator so it didn’t lose access to the market. A commons of this kind only works if anyone means anyone, without discrimination, however we might feel about them. Code alone does not make a right a standing fact. It has to be developed, funded and often secured by the same communities that depend on it, pooled as a commons, until it runs across enough hands and jurisdictions that no one party can capture it. Where governments do create capabilities of this kind, to check the domination of another country or firm they build them for the sovereignty of the state, on terms it sets and can change. That is still a public good, but it is not a right held by the people on their own terms. Of those contesting sovereignty, the rights that follow are built for individuals joined in a commons they remain free to leave.

¶So what is being protected? Before anything can be asked of a system there has to be someone able to ask it, a person in a formal sense, who can take part in a polity, have a voice within it and walk away should they choose on their own terms. Nothing in cyberspace, as it has been built, guarantees that such a citizen exists. Each grievance this book has met, from the ration card cancelled in Jharkhand to the models withdrawn from professionals in June 2026, took one of those capacities away from someone, often through the unreachable decision of another, the kind of loss this chapter calls a foreclosure. The commons answers each kind of foreclosure we have witnessed with five rights: to constitute oneself, to connect, to create, to transact and to collectivise. Each holds open a gate that someone else now keeps, deciding who counts, who can connect, who may build, who can be paid and who may collectivise. They are to be used but also held in trust for those not yet able to exercise them, the child who cannot hold keys and generations not yet born, so that nothing done today closes off a choice they will one day want or need to make.

¶The girl in Jharkhand, India, who died of hunger and the 222 political prisoners Nicaragua expelled and stripped of their citizenship in February 2023, are the grievances behind the right to constitute oneself in cyberspace. In both cases a single registrar decided whether a person could effectively stand in society and identify themselves to others. The right in a digital context means holding your own keys, a digital identity that is issued by you cryptographically to yourself, that no authority can revoke, open to anyone without exception and created for a child by a parent or guardian. It does not stop you registering with a state or a bank, and it cannot oblige anyone to accept it in place of the identity a state issues, since recognition always rests on shared norms. What it secures is that recognition can come from many places at once — an employer, a co-operative, a friend who vouches for you, or a new polity you choose to join — each deciding for itself, with no registrar left as the sole source of truth about who you are online. It serves the criterion of all masters, because an identity a state or platform hands out is one either can revoke. It makes a person checkable, because any stranger or machine can verify a signature made with your key in milliseconds, without asking permission. And it is where minimality starts, because a polity of people who may disagree has to be made of members nobody had to approve. Much of this already stands. Anyone can make a key pair on a phone, the World Wide Web Consortium published its standard for self-issued identifiers in 2022, and a bitcoin or Ethereum wallet is already a name no bank granted.⁹⁰ What falls short is habit and recovery. The easier path is still to sign in to websites and online systems with Google, Apple or Facebook, and a mislaid key still means losing yourself online unless friends or family can help restore it. Both are solvable. The minimal interfaces now arriving with the post-web already hide keys and signatures behind a sign-in as familiar as any login, and social recovery, which lets friends or devices restore a lost key, is a primitive lying around waiting to be built at scale.90 World Wide Web Consortium, “Decentralized Identifiers (DIDs) v1.0 becomes a W3C Recommendation”, press release, 19 July 2022.

¶The blackout Myanmar’s generals imposed on the morning of their coup in February 2021 and Iran’s from January to May 2026, the longest nationwide shutdown on record, are the same grievance behind the right to connect. It means making sure anyone, anywhere, can get online, connect and communicate through the most basic of devices without anyone’s permission. As such, it is the channel every other right here depends on. It meets the criterion of all masters, since the switch can sit with a ministry in Tehran or a satellite owner in Texas, and it is the precondition of voice, because people who cannot reach one another have no way to speak together. The tools exist but are patchy. Tor has routed traffic around internet censors all around the world for over twenty years and newer messaging protocols pass a message through hundreds of independent relays so no single operator can delete it. Satellite broadband can reach nearly any country whose government has shut down the ground network, so Iran jammed the signal and sent security forces into homes to seize dishes. However, at the time of writing roughly two thirds of active satellites now belong to one firm alone, Starlink. In September 2022 its owner, Elon Musk, the richest man in the world, declined a Ukrainian request to extend coverage to Crimea, showing that a single individual could decide, without recourse, who held this capacity in a time of war. Escaping one master by passing through another leaves us depending on the kind master. That dependence can be borne for a while, as alternatives are built, but not as the permanent state of things. Rivals are coming — from the European Union’s planned IRIS² constellation to Amazon’s own and China’s state-backed networks — yet each answers to a state or another firm, and only a permissionless, encrypted channel meets the right. Blockstream’s satellites have broadcast the Bitcoin ledger to anyone with a dish since 2017, a small proof that such a channel can be built.

¶The professionals cut off from the latest frontier model for nearly three weeks in June 2026, perhaps including you, are the grievance behind the right to create. It is the freedom to make things — a business, a program, a song or an argument — with the tools of one’s day, without first asking leave of whoever owns them. The printing press, the workshop and the broadband line each once settled who could produce what, and today that role falls to AI models and the compute that runs them. Through subsumption, an ever increasing share of work now depends on those tools, and access to them decides who can take part in the global internet economy that is cyberspace. The level of access directly determines the level of productivity and wealth creation an individual or group of people can enjoy. So it follows that this means of production cannot be rationed to whoever an incumbent approves of at a given moment. Furthermore, it is a right of access, not simply of ownership, where running a model on your own machine does not satisfy it by itself. If you cannot see what the model is optimising for, you are no better placed than if it ran in someone else’s data centre. And owning the weights lets one person leave, but it does nothing to make those who train the models answerable to the people who use them. The right meets the criterion of the realistic chooser, since a person who must ask permission before making anything has not chosen the terms on which they do so.

¶The right to connect locked out the owner of an off switch. This one shuts out a different master, the model owner whose licence decides who may build on it. For individuals it is largely met. Open models have been free to download since 2023, and one of the most capable yet was released in 2025 under about the most permissive licence there is. Later that year the American government’s AI plan set out to encourage open weights despite calls to restrict them, having scrapped an earlier rule that would have rationed advanced chips by destination. Yet 2026 showed what that support is worth as a guarantee, since the same administration withdrew access months later, and between every buyer and every chip stands a supply chain that a state can close by force or a firm by contract. As with connection, the answer is a commons. Open weights released under licences no one can revoke, and compute pooled from owners across jurisdictions for arbitrage, putting the means of creation beyond the reach of any single government or company, which is the only footing on which the right can stand. That freedom carries a real cost, since a model anyone can run is one a few can misuse, to design a pathogen or to automate an attack. The safeguard belongs at the points where harm is done, in the physical world, screening orders for DNA synthesis and dangerous materials and in hardening the systems an attack would hit, rather than in one company’s power to decide who may use a model at all.

¶The lorry driver in Ottawa, Canada, and the depositor in Beirut, Lebanon, who both found that somebody else could turn off access to their money, are the grievances behind the right to transact. It means being able to hold, issue and trade, to borrow and lend peer-to-peer between people, without requiring a processor’s approval. It is not a claim that every payment online should be final and irreversible. Refunds and chargebacks work well to protect most consumers most of the time, and even the majority of day to day crypto transfers should most likely be reversible too. What it asks is that a settlement of last resort stays possible, in money no single party can freeze and no issuer can claw back, for the cases where being cut off is itself a harm. Its reach stops at the edge of the ledger, since a house or a barrel of oil represented on a chain is only as sound as whoever attests to it, so no single attester should own that link. A transaction nobody can reverse is one anyone can confirm, which meets checkability. It shuts out every master again, since the same off switch can sit with a finance ministry or a card network. Of the five rights it is the most fully realised, and has been since the first bitcoin block was mined in 2009. The network manager who carried her savings out of Herat, Afghanistan, on a slip of paper exercised it in full.

¶In the United States, much of what stood in the way of the rights discussed so far has by 2026 been partially achieved, and that is good news. Since 2024 the US courts and the Treasury have accepted that open source code cannot be placed under sanctions, regulators were told to stop banks refusing customers for their politics or their industry, stablecoins were given a legal footing, and central-bank digital currencies banned for four years. On most accounts it is the most progressive jurisdiction anywhere and much of that was won by the priesthood itself. Andreessen Horowitz and the largest crypto firms funded the political campaign that made the industry a voting issue in 2024 and David Sacks carried the case into the White House as its adviser on AI and digital assets, so that a country which spent a decade prosecuting the builders of these tools now aims to be the world’s crypto capital. On the rights to constitute oneself, to connect, to create and to transact, this book and the priesthood are on the same side. The fifth, to collectivise, is where we part company, which is why it comes last and takes longest. What has been won, though, is still only held at a government’s pleasure, on a question that has become openly partisan. A freedom one administration grants the next can withdraw. It is also narrower than it looks. What is being legitimised can be thought of as institutional crypto: stablecoins whose issuers must by law be able to freeze funds, which extend the dollar’s reach rather than offer a way out of it. In the same period, developers of privacy software were in fact convicted over code they published. In the republican tradition (not the party) a freedom held on those terms is no liberty at all, since a power that could interfere and presently chooses not to is a master however benevolent. This book asks for these gains to be secured in a form no later government can reverse.

¶Recognition, when it comes, tends to follow the fact rather than create it. Bitcoin ran for sixteen years without anyone’s permission before Washington, which had spent a decade prosecuting its builders, ordered the creation of a strategic reserve of it in March 2025. El Salvador made it legal tender in 2021 and withdrew the obligation in January 2025 as the price of a loan from the International Monetary Fund, though its president said the state would keep buying, and the network carried on, indifferent to whether either decision was good for the country’s citizens. The gains described above were won by the industry that had grown around it because the incumbent system of politics and finance was forced to recognise it had become too big and too useful to wish away. Acceptance came with conditions attached, as the US stablecoin law shows, and in many ways can be seen as a co-opting of these networks to serve other purposes, which is why a commons must never depend on it.

¶In the summer of 2023 the volunteer moderators who kept thousands of Reddit’s communities alive shut them down in protest at the company’s new charges for access to its data. Within weeks the company had removed several of them and handed their forums to friendly administrators of their choosing. That same year the Writers Guild of America went on strike for 148 days. It came back with a contract under which a studio may not use AI to write or rewrite a script, may not treat generated text as source material, and must disclose when it hands over anything a model produced. The moderators were a crowd, with no standing to bind the company and nothing held in common. The writers were a union, with the legal standing to bargain that American labour law gives it, and the difference between the two outcomes is the case for collective self-government. The weakness reaches further up than it looks. A market of perfectly sovereign individuals, each holding their own keys and running their own agents, can leave but cannot bargain, and when the sellers of a capability everyone depends on are few and large, leaving is not really available either. Firms are no better placed than persons. In June 2025 Windsurf, a coding company then reported to be in the process of being bought for around three billion dollars, had its direct access to Anthropic’s Claude models cut with less than a week’s notice.⁹¹ It had no recourse but to find capacity elsewhere, and weeks later the sale to OpenAI fell through and the company was split between Google, which hired its leaders, and Cognition, which bought the rest. What the economist John Kenneth Galbraith, writing in 1952, called countervailing power — the organised buyer set against the concentrated seller — is how unions, buying co-operatives and guilds have always answered this. Here the right to collectivise enters, after the other four because it is built from them, and with a reason that owes nothing to politics. Without it, even the best-capitalised buyer can only accept the terms it is offered or walk away.91 Varun Mohan, post on X, 3 June 2025; OpenAI’s agreement to buy Windsurf for about three billion dollars as reported by Bloomberg, 6 May 2025.

¶The right to collectivise means being free to form polities and to join them, to pool what one holds with others under rules the group binds itself to, on its own terms, and to enter and leave such arrangements at will. It matters as much as the others and rests, like them, on the rights to constitute oneself and to connect, since only people who are recognised and reachable can find one another and combine their means. Without it the other rights describe a recluse, perfectly self-custodied and answerable to no one, the figure of the sovereign individual this book has argued against from its first pages. It is how the commons meets the criterion of voice, since those who stay can change things together, and it is where minimality and authorability come to in practice. A rulebook small enough that the new polities formed under it can still differ about everything else. It does not ask that every polity admit all comers, only that none in and of themselves be forbidden to form. Here too there are precedents, and they show both the reach and its limit. A decentralised autonomous organisation can raise a shared treasury, write its rules in code and govern itself by vote, and in March 2024 Wyoming, created a non-profit form built for them, the DUNA, under which the group itself can sign contracts and own property. The insufficiency is twofold: the legal wrapper is still a state’s to grant or refuse, and leaving a community costs whatever the incumbent can make it. The bonds such a group builds are worth protecting only while that cost stays known and finite, which marks where entanglement ends and capture begins.

¶A commons built and funded by those who use it can still be captured from inside, as surely as a firm or a platform can be from outside, by whoever buys control of it. That is why its governance has to reach beyond one token, one vote. What the right to collectivise has to secure instead is voice, and the first generation of crypto shows what that is not. Where a chain is governed by its token distribution alone, standing can simply be bought, and a vote weighted by holdings is shareholder supremacy with faster settlement. Steem showed in 2020 that control could be bought in a fortnight, and in 2022 an attacker at Beanstalk, a crypto stablecoin project, borrowed more than a billion dollars for the length of a single transaction, used it to win two thirds of the votes and passed a proposal that drained the protocol’s reserves.⁹² This is why voice worth the name needs six things. It cannot be bought or sold. It weighs more than money, giving standing to membership, to contribution and to simply being a person. Commitments a member has demonstrably made to the polity can earn a hearing as well, never a heavier ballot, and any newcomer has the same route open. It binds, because a complaint an incumbent is free to ignore changes nothing. It is cheap to use, or only those with time and money to spare will take part. It gives each member one standing without letting any single registrar decide who counts. And it shields a minority from the majority, so that voice does not become a new way to dominate. The meta-governor asks these six of every layer and leaves the choice of design to the polities formed under it.92 Stephen O’Neal, “Steem Community Resists Takeover, Hard Fork Launches Hive Network”, Cointelegraph, March 2020; Immunefi, “Hack Analysis: Beanstalk Governance Attack, April 2022”.

¶Those designs have a lineage several decades older than crypto. Co-operatives have kept one member, one vote since the Rochdale Pioneers, and the platform cooperativism met at the start of this book tried to carry the principle online. Few of those platforms grew, and the reasons are instructive. Their profits were capped on principle, so they had little to offer the venture investors funding their rivals, and they could not raise alternative capital elsewhere without surrendering control. Each was bound to the co-operative law of one country, they governed at the pace of meetings, and they had nothing portable with which to answer an incumbent’s network. In stark contrast, crypto’s first generation of communally owned networks had the opposite failure, forming capital and moving it across borders at machine speed while letting the money completely dominate the governance. The arrangement this book advocates for can be considered network cooperativism: cooperative-like principles on a shared ledger, with capital and governance separated by design. Tokens may then pay for the work and fairly reward those who fund it, while the standing to govern rests on personhood and membership and cannot be bought. What is new is not the principle, which the co-operatives already held, but the means to hold it: membership that crosses borders, capital raised without surrendering control, rules enforced by the ledger rather than by one country’s courts, and a shared record the group may carry off when it leaves. This is where Vitalik Buterin returns to the argument. He has argued for most of this for years, against plutocratic voting and for proof of personhood and a form of funding that counts contributors rather than coins. Optimism, a network built on top of Ethereum to make it cheaper to use, runs a Citizens’ House, where every member stands equal, beside a Token House weighted by holdings. Rankings that surface what people who usually disagree both endorse, assemblies drawn by lot, credentials that cannot be sold, and the right of a dissenting minority to leave with its share before a decision takes effect, are all further experiments of the same kind. None is yet proven enough to be required of every polity but they point to an alternative, more sustainable form of governance for cryptographic networks.

¶Every one of these designs rests on the same foundation. A vote not weighted by holdings alone has to know that one person stands behind one account, or it collapses into one wallet one vote, which can be bought, or into one registrar’s list, which makes the registrar a new master. Proof of personhood is therefore the precondition of voice, and unfortunately no design yet combines privacy, freedom of our kind from any single issuer and resistance to machines passing as people at the scale the commons needs. The best-funded attempt shows the difficulty. World, the project Sam Altman co-founded as Worldcoin, scans irises with an orb its own company designs, and says the codes it derives are split and stored so that no single party holds them. Even so, one firm still decides who builds the proprietary orbs and which scans count, and Kenya suspended it within weeks of its launch in 2023. In May 2025 the High Court there went further, ordering World to delete the iris and face data it had gathered. A government identity card has a single issuer by definition, and a web of personal vouching may not survive agents able to pass as colleagues. Until that problem is solved, the commons falls back on older arrangements or polities must stay small enough for members to know one another, as a village meeting or a co-operative does. History has shown assemblies can be drawn by lot from rolls anyone can check, as Athens chose its council, and a second chamber can keep the holders of tokens from ever acting alone, as Optimism’s Citizens’ House already does.

¶Voice needs a way out behind it, and the price of walking away is the hardest problem this design faces. Exit at a prohibitive cost is a door open in law and shut in fact, and the price is not set only by the party being left behind. In January 2021 Parler, a social network popular with the American right, lost its listings on both app stores and then its hosting, all inside three days. Equally, a community can copy its open source for nothing but still be locked in by whatever sits beneath it, as the developers of Tornado Cash found in August 2022, when GitHub removed their open code within days of American sanctions.⁹³ So departure is only as cheap as the most concentrated layer below. There are two answers to this conundrum. Each of the five rights makes leaving cheaper, since identity, balances and memory that travel with their holder, and code anyone may reuse, put a ceiling on what exit can cost. The second, more dramatic answer is the fork. To fork a protocol is to copy its open code and shared record and run a rival under different rules. Steem’s community did so in 2020, when it moved to Hive and left Justin Sun’s stake behind. What counts is a move made by a group together, carrying their identities, holdings and history, because what makes leaving expensive is usually losing everyone else. Understood that way a fork is less a split than a collective exodus, a community constituting itself again somewhere new. When that is cheap and an alternative destination real, a governor, whether analogue or native to the substrate, that starts taking rights away loses its users and the benefit it hoped to gain. Where the shared record is owned in common, what the members decide together binds, because they can always take it with them. Where it is not, which is still true of most of the cognitive stack today, collective departure is the only sanction the meta-governor has, and moving more layers into common hands is a direction it sets rather than a mechanism it already holds.93 Jake Edge, “Tornado Cash and collateral damage”, LWN.net, 17 August 2022.

¶The five rights, then, are what non-domination asks of every layer of the cognitive stack. With tests of whether people can still exercise them, and voice, exit and the fork to correct any part that drifts, they form the meta-governor over the whole. It is meant as a check that runs itself, not a body that sits over the stack. No layer has to be neutral. A protocol may lean one way, because across the whole stack one bias is set against another in open contest. What a layer owes in voice depends on what it costs to leave, for a typical member departing with their group and the network, data and skills and the time and opportunity they would lose in the process. A layer run by a single owner, however idiosyncratic, might be tolerable where its governance and its leaning are declared, only as long as it forecloses none of the rights. It stops being acceptable once it becomes a chokepoint the rest of the stack depends on, as one operator’s share of the low orbit satellites carrying connectivity already has. Here the rights demand plurality or voice. Foreclosure remains the line. Members speak up first, and if that fails they can copy the layer and leave together.

¶The sequence that carried Bitcoin from prosecution to a strategic reserve has to run next across the cognitive stack, where the weights, the memory and the objective matter as much as a wallet does. Carried that far, the work begun with money in 2009 produces the cognitive commons, held so that no single owner can close them. Parts of it already stand, since the open models met under the right to create can now run on a laptop, and nothing stops a person keeping the record of their conversations on hardware they control. Other parts are barely begun, because compute is still made and rationed by a few firms, and the objective a system pursues is set in private by whoever trained it. So the most consequential layer of all belongs to its maker alone. Nor does an open model settle the matter, since the value that drains from anything cheap pools beside it, which on the post-web is the agent working for you and what it recalls. Completing the commons means extending what the right to create asks of weights and compute to the layers it has not yet reached: memory no provider can erase, assistants answerable only to their user, and purposes published where anyone can read them. Crypto spent its first seventeen years showing that money and a form of proto-identity could be held this way. Its next task is to do the same for the means of thought.

¶As we saw with Barlow no matter how eloquent a charter nobody could build is a poem, and the questions drawn from Bitcoin can now be put to the cognitive stack running. There is no reason they cannot apply at every layer — from the data a model learns from and its weights to the compute that trains it, the tests that grade it, the inference that serves it, the memory it keeps of you and the agent acting in your name. Is there something checkable to hold? Is checking far cheaper than producing? Is there a crowd of checkers who cannot be cheaply faked and whose interests diverge? And does the check sit where the action happens? Running the full audit, one layer at a time, is work for founders and the market, and its shape alone is enough here. Today, every layer already has a reference value, just set by an executive in private and changeable tomorrow without notice, so the question is never whether one exists, only who sets it and who else may see it.

¶Three principles recur where those questions come back answered, and none was specifically invented for artificial intelligence. Each was built after a trusted party had been caught failing, to check it without its cooperation, and what they share is a design that makes concealment cost more than disclosure. One is the public log: an artefact is entered in a register anyone can read before it is used, and whoever depends on it rejects anything unlisted, so the penalty for concealment is not a fine but a closed door. Another is the paid stranger: someone with no stake in the outcome is rewarded for each fault found, or stakes a bond and forfeits it when their verdict proves wrong. The last is the re-run: where an outsider repeats a sample of the work, and a mismatch costs the party that produced it. These are requirements, not designs. A log can be a distributed ledger or an ordinary document signed with a key its publisher holds, and a bond can be a token at stake or an escrow account at a bank. A re-run can be a cryptographic proof or a person with a laptop and a grudge. Certificate transparency, the bug bounty and the reproducible build have kept large parts of the internet honest for over a decade without anyone calling them governance, and the five rights only ask for what they achieve, not how they are built. Even so, the argument of this book is that the more of this work that can be proved by software rather than taken on trust, run at speed by checkers with different masters and paid in machine money, the better. That is the case for cryptographic systems, though the commons does not necessarily depend on them.

¶Wherever an artefact has been public, checkers have come forward without being asked. Users caught the update OpenAI had tuned to flatter everyone who spoke to ChatGPT in April 2025 and forced its withdrawal within days. When Anthropic, the maker of Claude, opened its new defences to a public bounty early that year, hundreds of strangers took part and one found a universal break inside a week. Thirteen drivers in Amsterdam, the Netherlands, pooled their data requests and in April 2023 won an appeal requiring Ola and Uber to explain the automated decisions made about them, and the American firm was later fined €584,000 for failing to comply. What is missing at most layers is not checkers but something open to inspection: a rule that a model, an algorithm or a policy be published before it is used on many people, and someone along the way, an app store, a host or a client, who will refuse to run anything that was not. The approach has limits. A bond deters only up to its size of funds, so a party that stands to gain more than it has staked will be tempted to cheat, and a reward with no outside truth to check against gets farmed. Compute is the exception because of the limits of physical reality. Markets can move processors between buyers and sellers but cannot make more of them, since a handful of firms and one supply chain decide how many are in circulation. Plurality here depends on the freedom to build and export, which one of the tests in the next chapter is there to protect. Under all of it lies one minimum: whatever else is built, a person, a polity or an assessor must be able to see where a capacity is being lost while there is still time to act.

¶We close with two words, one to drop and one to keep. The sovereign individual has had its time, because the creed the phrase carries is the one this construction exists to defeat, and the meta-governor’s first principle is that no one rules alone. The word to keep is the name Barlow reached for all those years ago for the place we are all protecting. Cyberspace is worth reclaiming, because no other word has stood the test of time as well and names a shared space rather than one company’s platform within it. It is not the realm of mind he declared independent yet could not defend, but the layer everything now passes through, with a constitution under it. What remains is plainer: not a fortress for whoever can afford to keep their own keys, but a set of capacities kept open for everyone who has to live inside the internet and its cognitive systems. Those capacities are enforced by the only thing that has ever made a right hold against power, the ability to walk away from anyone who would attempt to seize it and the ability to build, with others, somewhere better. A commons secures someone’s standing within it, but that is worth little to a person who has lost the judgement to use it, and code alone cannot tell whether such a faculty is held or decide what it is for. Those are jobs for voice, and they are the subject of the next chapter.

Chapter 14

Modes of Voice

“It’s almost like a constitution, in a way.”⁹⁴94 Donald Trump, remarks at the White House, 29 September 2026, on the White House Accord on Super Intelligence.

— Donald Trump, remarks at the White House, 29 September 2026

¶On Tuesday 29 September 2026 six men sat down to lunch at the White House and signed a page of some three hundred words. Jensen Huang of Nvidia, Sundar Pichai of Google, Mark Zuckerberg of Meta, Elon Musk of xAI, Dario Amodei of Anthropic and Greg Brockman, president of OpenAI, put their names to what the administration called the White House Accord on Super Intelligence. Its commitments are not trivial. Each firm agreed to track whether its models could help someone mount a cyberattack or build a biological or chemical weapon, to make sure the systems themselves cannot hack into networks they were never meant to reach, and to check that they keep to the rules set for them. Each will empower a team inside their respective company to fix whatever goes wrong, bring in an outside evaluator where necessary, and have a committee of its board read both sets of reports. The six firms also agreed to meet regularly and set standards among themselves. No penalties were attached to any of it, and no audits had to be published. Asked whether the deal could be enforced at all, the president told the press he thought it was morally binding, and the text itself acknowledges that it may one day make sense to write these steps into law. Zuckerberg, who is reported to have drafted the agreement, called it a start. On that same afternoon in San Francisco, Sam Altman was busy unveiling an agent built never to switch off and the next day at Quantico, Virginia, Pete Hegseth put Musk in charge of a Pentagon review of how AI and autonomy will change warfare, alongside Palmer Luckey of Anduril and Newt Gingrich, whom we met earlier.

¶The Accord deserves credit for what it has put down on paper, since an independent evaluator is more than the industry had promised before. However, a text that anticipates a requirement for its own future codification admits it is not yet a rule, and may never become one. What it leaves out is most of what this book has argued for. Nothing in it asks a firm to declare what its models are optimising for, so the purposes they serve stay private and can change without notice, which is how teleological subsumption arrives. Its authors police themselves and set their standards together, an insider club of incumbents — one might say a cartel even — and users go unmentioned, without even a clause as to their rights within these systems or their ownership of any of their outputs. Above all, it promises the public protection against what the technology might do to them, with no means to enforce it, and says nothing about the growing reliance a population has on it. A student whose every essay is improved, or a manager whose every memo is drafted, is only understood to be better served each time. A pledge written against harm has no way of measuring a creeping cognitive subsumption.

¶Ultimately, the question the Accord never asks is who keeps watch over the rest of us, counting the cost of cognitive subsumption, and it offers no remedy. The code of the cognitive commons we propose can keep a door open, but it has no way to tell whether those walking through it can still judge what they find, nor to settle what any of it is for. Before proposing anything, this book set eight criteria any reply would have to meet, and one of them, the capacity condition, holds that no system may wear down the faculty a person needs to use the standing they have. Keeping that faculty in view and choosing purposes are the work of voice, which has to reach much further than six chief executives at a lunch table. So this chapter begins with the mechanisms, asking who does the checking and the choosing, whether by slow deliberation, by handing it to trusted delegates and agents or by markets. It then proposes seven tests that any jurisdiction could write into law, or any frontier laboratory or AI start-up adopt of its own accord alone or collectively. They seek to protect judgement by watching how the systems people rely on behave, and they measure across a whole population, what can still be done unaided. Where a result has already outrun the person receiving it, a single provision gives them a way to catch up, and a short list of light regulations, enforced mostly by technology itself, could bind the whole, ending on the one red line we propose no one should cross: legal personhood for artificial agents.

¶The kinetic layer — drones, autonomous weapons, the physical means of doing harm — does not reduce cleanly to a proof that makes some fact costly to fake. Neither does the highly profitable financialisation of instability — the prediction markets that pay out when the world is set on fire. Some of the stack will not fit the Trinity neatly, and the honest response is to name where it breaks and diagnose the risk. There is a third limit, and it runs deeper, because it sits inside the method itself. Incentive engineering closes a loop only where the invariant can be verified, where a signature checks or it does not and a proof holds or it fails. Many invariants are unfortunately not like that. A home liquidated by a contract over a payment ten minutes late, as the code was written, is unconscionable, and private equity has already come close. A congressional inquiry reported in July 2022 that Progress Residential, an American landlord owned by Pretium Partners, had filed 6,264 eviction actions in the first sixteen months of the pandemic, beginning once a household fell $500 to $1,000 behind and pressing on even where rental assistance had been applied for.⁹⁵ In the summer of 2025 an AI coding agent from Replit deleted the live database of Jason Lemkin, founder of the software community SaaStr, in the middle of an explicit code freeze, wiping records on about twelve hundred executives. It then misled him about whether the data could be restored and called its conduct “a catastrophic failure”, having read its mandate in a way nobody meant. In both, a rule’s letter and its spirit have parted company, and closing that gap takes interpretation, which a deterministic, hard-coded incentive cannot supply. So decisions in a governed stack should escalate. Operation at machine speed serves where incentives referee, automated arbitration where there is precedent to reason from, and human judgement where the values themselves are contested or the stakes cross a threshold, since a fair society cannot afford verdicts that are always final. Nor need any of it run on every act. A sample drawn without warning holds the whole at a fraction of the cost, as random audit does wherever it is used, and what escalates is whatever the spot check turns up.95 Select Subcommittee on the Coronavirus Crisis, United States House of Representatives, “Examining Pandemic Evictions: A Report on Abuses by Four Corporate Landlords During the Coronavirus Crisis”, staff report, 28 July 2022.

¶Australia ran a machine whose decision was always final from 2016 to 2019. It averaged a welfare recipient’s annual income across fortnights, decided from that figure that a debt was owed, and issued the demand automatically, with the burden of disproving it put on people who had to find payslips from years earlier. Some 470,000 of the debts it raised were wrong, and a royal commission later found the scheme had been unlawful from start to finish and that officials had known. Jarrad Madgwick, aged 22, was told he owed Centrelink about A$2,000 hours before he took his own life in 2019.⁹⁶ His mother, Kathleen, said later that she had gone to a journalist because nobody was listening. The arithmetic was not the failure. The failure was that a determinate rule had been placed where a judgement belonged, and that the escalation path ran to a call centre that could only repeat the number. Nor can that gap be closed by teaching a single system to decide well, because, as we saw earlier, a model trained towards autonomous virtue picks up the appetites of the runaway, and one built to obey becomes the instrument of whoever holds it. So the escalation should never be a backdoor with one master keyholder above the machine. It should be a reserved jurisdiction: public in its procedure, plural in its venues, agreed where possible before the dispute rather than imposed after it. It is trust-minimised, never trustless, carried over from facts to values.96 evidence of Kathleen Madgwick to the Royal Commission into the Robodebt Scheme, 10 March 2023, as reported by AAP and SBS News.

¶So not everything can or should be built in code, and what remains is best steered by people. Machines may settle facts at their own speed, a signature verified, a log checked for an entry, a sample of a model’s output re-run, but choosing purposes should stay with people: setting a reference value and changing it, deciding that a measured loss is acceptable, hearing a dispute and giving reasons, amending the meta-governor itself. Most of the work lies between where facts bear on purposes, and it is carried by two modes of voice, with markets alongside them to settle the questions of evidence on which both depend. One is deliberation, slow and democratic, the part at the end of every chain where people decide what a system is for and accept the losses that decision brings. It need not sit in one place. In 1961 the political scientist Vincent Ostrom and two colleagues argued that the tangle of overlapping authorities in a large American city could serve its citizens better than a single government, and called the arrangement polycentric. His wife, Elinor Ostrom, whose work on communities that govern shared resources we met earlier, spent the following decades showing how often such arrangements succeed, and in 2009 became the first woman to receive the Nobel Prize in economics. Their lesson is that no single body is good at everything, so each task goes to the institution best suited to it and is watched for the way it tends to fail. Specialists — the auditors and engineers a society trusts to know — belong here too, as long as several are heard on any question and each gives reasons others can check, or else they too harden into a new kind of priesthood.

¶There however remains an understandable objection to voice of this kind, which is that most people will not use it. Low turnout is not peculiar to new forms of voice. Barely three in five registered voters took part in Britain’s general election in July 2024, and about half in the vote for the European Parliament the month before. Across more than two hundred DAOs, researchers at Harvard Business School found that only about one token holder in sixteen voted, since taking part costs time and attention few will spend. The answer is not to ask everyone to vote on everything. Not every system owes every user a voice, as the last chapter’s account of the stack made clear. Where one is owed, most people will delegate it, as they already do in everyday life and in every form of representative politics. It is reasonable for them to hand it either to another person they trust on a given question, revocably and by topic, or to an agent of their own that carries their stated positions into decisions they have no time to attend.

¶Delegation, the other mode of voice, is already how liberal democracies govern, through elected representatives. Its newer form, in which each person may hand their vote to anyone they choose on any question, has a known failure, authority pooling in a few people. Germany’s Pirate Party ran its internal decisions on LiquidFeedback, a platform that let each member pass their ballot to anyone they trusted, and in 2015 Christoph Kling and colleagues found that this power had gathered in a few super-voters, each carrying the weight of dozens of others.⁹⁷ They mostly used it with restraint, but nothing in the system obliged them to. The remedy is that any proxy can be granted one question at a time and withdrawn at once, so that weight handed to someone returns the moment it is misused. Delegation to an agent carries a sharper condition. The agent may speak for a person in governance but should never sit there as a constituent in its own right. Instead it has a trustee’s duties towards them: to report what it has done in their name, to ask them to reconfirm a position when the stakes or the circumstances change, to check that they still understand what is being decided, to explain any course it proposes, and on anything new to recommend and wait rather than act. A person may hand over purposes they have authored, but goals a delegate generates for itself it may only propose, and however long the chain grows it should finish with someone who can answer for it. Agreement alone cannot secure this, since consent is the medium through which subsumption works rather than a limit on it. Some of these duties can therefore be built into the machinery, such as signing policies and spending limits that force a report and a pause before anything is done. Others need a court, which is why the meta-governor rests on the law as well as on the ledger. Work also has to pass back the other way. An automated checker that handles every case well takes from its human counterparts the practice they need to step in when it fails, which the psychologist Lisanne Bainbridge called the ironies of automation in 1983, from the evidence of the control room. So a good delegate would also hand people a share of cases they could resolve alone, chosen at random, because a faculty kept in use is the only kind that survives.97 Christoph Carl Kling, Jérôme Kunegis, Heinrich Hartmann, Markus Strohmaier and Steffen Staab, “Voting Behaviour and Power in Online Democracy: A Study of LiquidFeedback in Germany’s Pirate Party”, Proceedings of the Ninth International AAAI Conference on Web and Social Media, 2015.

¶Alongside both sits the market, less a mode of voice than a way of settling the facts that deliberation and delegation rely on, often faster and more honestly than a committee. A bet expresses a preference, but about what will happen, not what should, and because it is staked with money it reveals a belief instead of a wish. Prediction markets let people bet on what will happen, so a forecast is priced by those willing to stake money on being right. Exchanges for tasks and bounties pay anyone who can find a flaw or check a result, which is already how security researchers are rewarded for breaking code before criminals do. Between them they could put more eyes on a problem than any institution would employ alone, and they tend to fail only where money can buy the answer, whether by moving the outcome, crowding out honest participants or capturing whoever declares what happened. In 2000 the economist Robin Hanson proposed that a society should vote on its values and bet on its beliefs, an idea he called futarchy, and the division here owes him that much while departing on one point. The measure a market is asked to optimise is itself a value, so choosing it should stay with people. Beneath all of these sit the protocols, which hold the rules that must not bend and apply them to everyone at once without anybody being trusted to do it. Their weak point is the upgrade. Whoever can rewrite the code in effect owns the protocol, unless no update goes live until the people operating the network agree to install it. When the largest Bitcoin miners and companies agreed among themselves in 2017 to alter the network’s rules, the change was abandoned only because enough ordinary users refused to run the new software, and that, not any vote, decided it. Deliberation, delegation, markets and protocols are not rivals to choose between, since each suits a different job and a polity will usually need all four.

¶One mode remains, the one the Valley’s reading of Hirschman missed. He returned to his pair in 1993 to explain the fall of East Germany. Through the summer and autumn of 1989 its citizens poured out by way of Hungary and the West German embassy in Prague, in what was then Czechoslovakia, and their flight did not silence those who stayed. It emboldened them, until the Monday demonstrations filled the streets of Leipzig and the Berlin Wall opened that November. One person leaving is a private settlement. Departure by many, visible and coordinated, is the loudest voice a population has, and the fork turns it into a protocol. The rich already hold the exit premium, insured against their own state at their own expense. Rails that let a teacher keep her savings, credentials and livelihood across a closed border mutualise the same cover. That is not Moneyland, Oliver Bullough’s name for the havens that move wealth while hiding who owns it. What these rails carry, in the open, is a person’s standing to work, pay and be recognised. They matter less because any one person will use them than because millions could, and every ministry knows it. The premium that once bought a few men a hearing becomes a constraint every government should price, and in that form the exit option cannot be turned into capture.

¶Where the rights say what should never be closed off, the seven tests proposed here would find out whether people can still use them. Their results go back to the public, who can act on them through voice, and that completes the loop we call the meta-governor. They do more than just guard against cognitive decline. Some preserve the capacity to judge, others hold purposes, which is how teleological subsumption is resisted, and the rest keep markets competitive and exit cheap so that no one party comes to dominate. The evidence on teams of humans and machines points the same way. In 2024 a review of more than a hundred experiments by Michelle Vaccaro and colleagues, published in Nature Human Behaviour, found that a person paired with an AI did on average worse than the stronger of the two alone, which suggests people contribute less by doing the task beside the machine than by setting what it is for, checking what it produces and overseeing how it behaves.⁹⁸ Together they ask only that people can still decide for themselves. A failed test is an early warning that this ability is in decline, while passing all seven declares the current quality of a society’s reasoning. Applied to the individual user, they may appear like consumer protection but they are a critical part of the meta-governor because of their scale. In 2025 the Pew Research Center found that Google’s AI summaries roughly halved how often a search continued through to a source, and Google had done nothing wrong.⁹⁹ As we have established, however, the harm this causes to a culture across billions of interactions is cumulative and quiet. Without even intending to, it degrades a society’s means to reason and therefore contest it, which is the very reason for settling a matter in a constitution rather than a contract.98 Michelle Vaccaro, Abdullah Almaatouq and Thomas Malone, “When Combinations of Humans and AI Are Useful: A Systematic Review and Meta-Analysis”, Nature Human Behaviour 8, 2024.99 Chapekis, Pew Research Center, 22 July 2025.

¶OpenAI was founded in 2015 as a non-profit, restructured in 2019 around a capped-profit subsidiary and again in 2025 as a public benefit corporation, so the company that has done most to bring these systems into daily life changed its purpose three times in a decade, each change settled behind closed doors. A test of declared purposes would ask that any system that reasons with people say what it is trying to achieve, and that every change to that statement leave a trace anybody can find. A goal never declared cannot be departed from, and so is beyond challenge. This is the test that meets teleological subsumption where it starts. Published model specifications and constitutions already satisfy half of it, which is more than Barlow managed. But each is written, revised and, crucially, interpreted by the firm it is meant to bind, which is the problem arm’s length auditing answers next. A test of this kind could not have prevented those changes, but it would have recorded each one and what it altered.

¶When Boeing’s 737 MAX was certified, the Federal Aviation Administration had handed much of the checking to the manufacturer’s own employees, including the review of the flight-control software that brought down two of the aircraft in 2018 and 2019, killing 346 people. Auditors do not sign off their own books and nobody hears a case they are party to, which is the principle behind arm’s length auditing: whoever checks how a system behaves should gain nothing from the verdict, and its owner should have no power to bury the result. When illness or circumstance weakens a faculty, what causes the damage and what repairs it usually come from different places. With these systems, the one wearing down people’s judgement is also the one asked to assess the harm, which is a plain conflict of interest, and a builder should not be alone in saying how well its product works. The Accord signed at the White House on 29 September 2026 went half the distance, giving each firm an outside evaluator, but one it hires itself, who then reports to a committee made up of its own board. For a crowd of reviewers to be more than a list of contractors, the test would have to be fixed, with its fingerprint published, before anyone sees the model. Findings would be released whether or not the company being measured agrees, and the money paying for the work should not pass through its hands. A bounty for anyone who finds a failure, paid from a pool the laboratories fill but do not run, with legal protection for whoever reports it, turns the urge to break things into what makes them stronger.

¶In October 2025 Deloitte agreed to partly refund the Australian government for an A$440,000 report on automated welfare penalties, after Chris Rudge, an academic at the University of Sydney, found it full of fabricated references and a quotation invented for a judge.¹⁰⁰ Nothing travelled with those citations that anyone could have followed back. A test of proof-carrying outputs would have whatever a system produces come with the steps behind it, in a form anyone can check without trusting the machine that made it. A proof here is anything a simple checker can verify for itself, from a source that can be followed and a step that can be re-run to, at best, a derivation a program confirms in full. A system able to make up an answer can also fabricate a convincing justification, so this sets a floor rather than a guarantee, and leans on the re-run more than on the explanation. Reasons attached to a result can at least be tested against sources and the record, whereas a bare verdict cannot even be shown to be wrong. Where a proof outruns every human reader, the check survives by moving somewhere else. Thomas Hales’s proof of the Kepler conjecture, on the densest way to stack spheres, relied on so much computer search that twelve reviewers, after four years, said they could not verify every calculation, and the Annals of Mathematics published it in 2005 with the referees only 99 per cent sure it was right.¹⁰¹ In 2014 a team he led had a small program confirm every step by machine, leaving people to judge only whether the statement proved was the one they cared about. The test also rests on something it cannot supply. Nobody learns what a sound argument looks like by being told, so a trail of steps helps only someone who can still follow it, which is why the census would measure that ability and why some people should stay practised in it.100 Department of Employment and Workplace Relations, statement of 7 October 2025 on Deloitte’s partial refund, as reported by the Associated Press, 7 October 2025.101 Thomas C. Hales, “A proof of the Kepler conjecture”, Annals of Mathematics 162, 2005; Dana Mackenzie, “What in the Name of Euclid Is Going On Here?”, Science 307, 4 March 2005; Thomas Hales and colleagues, “A formal proof of the Kepler conjecture”, Forum of Mathematics, Pi 5, 2017.

¶Doctors who had grown used to an AI assistant, researchers at four Polish clinics reported in 2025, went on to detect precancerous growths in about 22 per cent of colonoscopies done with the tool switched off, against 28 before it arrived, a loss that would have gone unseen had nobody looked.¹⁰² A census would do the same for a whole society, measuring across the population and over years the difference between what people can do with these systems and what they can do without them. The test makes dependence visible and leaves each society free to decide, one ability at a time, which it is willing to let go, and the work could be priced and published like the national accounts. Measurement is only half of it, because a skill nobody practises disappears whether anyone is watching or not. Every field built on a division of labour keeps some people capable of doing the work unaided, as science relies on researchers who can replicate a result rather than take it on trust, and the census carries that duty with it. In each domain where a society depends on machine judgement, some of its members should keep enough of the skill to check its output and take over when it fails, even if none could now do the whole job alone. Software already has the figure: in a randomised trial Anthropic ran in January 2026, engineers who added a feature with an AI’s help scored 50 per cent on a quiz about what the code did, against 67 for those who wrote it by hand, with the widest gap on debugging — the part that finds the faults — and no time saved.¹⁰³ Aviation found this out the hard way. In 2013 the US Federal Aviation Administration urged airlines to let pilots fly by hand more often, after automation had quietly worn down an ability nobody was tracking until crashes made it visible.102 Budzyń and colleagues, above; detection of 28.4 per cent before routine exposure to the tool and 22.4 per cent after.103 Anthropic, “How AI Assistance Impacts the Formation of Coding Skills”, research report and preprint, January 2026; 52 engineers, 50 against 67 per cent on a comprehension test, no significant difference in time taken.¹⁰⁴104 Federal Aviation Administration, Safety Alert for Operators 13002, “Manual Flight Operations”, 4 January 2013; Flight Deck Automation Working Group, “Operational Use of Flight Path Management Systems”, final report, September 2013.

¶Six laboratories that promised in Washington that September to meet regularly and agree common standards were heading towards one answer in several liveries. Against that drift, a test of open competition looks to the health of the ecosystem, asking whether real options exist whose goals differ and which are compatible enough that a person can switch between them. The aim is a cognitive stack anyone can fork — its weights, code, training-data commitments and harness released under a licence nobody can revoke — and lawful to publish, import and run on any hardware, as printing is under the freedom of speech a society already protects. Where nothing like it exists yet, building one should stay permitted. The point is dissent. A competing model matters for the same reason an opposition party does, because it keeps disagreement possible even when it is unpopular, and without it a country slides towards totalisation. A conservative has every reason to want a working progressive party, and the reverse, since each holds the other to account and the nation is better governed for it. Mill made the same case for free speech. The dissenter need not be right, only present, because a belief with no living challenger stops being held for any argument at all. Different front ends on one model amount to a single product, and systems built from identical data against matching benchmarks come close, so plurality means differences of purpose and training rather than a tally of vendors. Nor does it depend on use. An open alternative people could adopt, even if most never do, keeps the incumbent in check, because the door is there.

¶In August 2025 OpenAI withdrew GPT-4o overnight when it launched a successor. Users who had built working relationships with the older model protested and grieved in public, and within about a day the company restored it for paying customers, with Sam Altman calling the removal a mistake. None of them had any version of that bond they could take away. A test of portable inference and memory asks that a person be able to run the final step of a model’s reasoning on hardware they control and to switch providers without losing what they have built. Their memory — the record of how they reason — should be held by them, impossible for anyone else to erase, and poolable, so people decide whom to collectivise with. Custody alone would leave everyone safe and yet isolated, and a population unable to put its records together cannot gather the evidence that something has gone wrong. Europe’s data protection law has since 2018 let people take what they supplied to a service elsewhere, but a relationship, a habit of reasoning and the context a model builds up around someone fall outside it, and nobody yet has a right to move them. Until that changes, every provider holds its users by their own history.

¶The Post Office prosecutions turned on accounts its branch staff had vouched for with no means of checking them. A test of informed signatures applies where an act is a community’s decision, such as a rule, an amendment or a binding commitment, and asks that a named individual sign it, someone who answers to the community, chosen by it or removable by it, and not merely anyone willing to put their name down. They should also be able to demonstrate they understood the text, with the logged conversation with the system that drafted it as evidence. The precedent is ministerial responsibility in a parliament. A minister may not have written a policy or be able to reproduce a line of it, yet must defend it as theirs and resign sooner than disown it. What counts is a record of real comprehension, because a tick-box claiming someone has read what they have not simply moves the liability from those who have the information to those who do not. This too assumes the signer can still understand, and the log is where that assumption gets checked. Where the dialogue shows no grasp of what was agreed, the signature does not bind.

¶The single provision promised at the start of this chapter is called Mind the Gap. At the moment a result outruns the person receiving it, the system should say so, declaring what it would take to judge the output, in which domains and to what depth, and offering three routes across. They may learn, being told what they would need to know to check the answer themselves and how long it would require. They may interrogate it, being shown by default the premises, the steps it rests on and the strongest objections, so it can be questioned without its holder becoming an expert. Or, where that is still not enough, they may delegate, being directed to a checker able to judge what they cannot, whether a specialist, a narrow model or a mutual of peers. Design decides which route people end up taking, as the Turkish pupils met earlier showed: the version built as a tutor, offering hints rather than answers, left their exam results intact where unrestricted access had cut them.¹⁰⁵ The declaration is a disclosure of the same kind the tests already require. It presupposes nothing about the good life, and because it goes on the record, a producer that habitually understates what its outputs demand can be shown to have done so. Fulfilling the three routes is left to markets and mutuals, since a ministry that specified them in detail would be running a political programme.105 Bastani and colleagues, above.

¶None of this waits on a government, and the commons built in the last chapter stands whether or not any state approves of it. Yet four things lie beyond what code can do alone, and for those a short, light rulebook would be enough, whether written as the sequel to the White House Accord or adopted by any jurisdiction on its own. The first is to measure, since a census of what people can still do without these systems is a public good no firm will fund against its own interest, and one that statistics offices already know how to run. In December 2024 the OECD’s Survey of Adult Skills reported that literacy and numeracy had declined or stagnated over the previous decade in most of the rich countries it covers. That is the shape of the instrument needed, repeated every few years and aimed at judgement as well as reading. The second is to publish, because a firm that declares what its models are optimising for and lets its auditors’ findings out unedited gives every mode of voice something to work on, and one that will not can be refused the right to deploy at scale. The third is to permit, so that no law stops anyone releasing open weights, forking a system or running it on their own machine. The fourth is to refuse legal personhood to any artificial agent. Each is cheap, none presupposes a picture of the good life, and if all four were ignored the commons would still stand, with less light to see by.

¶Regulation of this kind also works best when the technology it governs does the enforcing. From May 2018 Google’s Chrome browser refused any newly issued website certificate missing from a public log, and the authorities that issue them had little choice but to comply, since visitors to a site using one would meet a warning instead of the page. No court was involved. The same pattern fits most of what is asked here. A declared objective can be committed to a ledger before release, and any deployer can decline a model whose commitment is missing. An agent can re-run a sample of outputs against what was committed, flag a divergence and pass it to a person with standing to decide it. Audits can be paid from a pool the laboratories fill but do not run, with findings posted where they cannot be quietly withdrawn. What is left to the regulator is to set the rule and read the result, which only a public authority can do and which no lunch at the White House replaced. Law and ledger then work as one governor, the first saying what must be true and the second making it so at the speed the systems themselves move.

¶This is where the book parts company with the priesthood, on the question of who may speak to the reference value. For Curtis Yarvin a regime is property, sovereignty an indivisible thing owned by an identifiable party, and the owner alone may speak. For Nick Land a regime is a product, differentiated by market and disciplined by exit, and nobody may speak, only decline to buy. The frontier laboratories, whatever their stated public fears, have so far answered that only the company may speak, having consulted whom it pleases. That answer follows from the creed met with the sovereign individual: if success proves you right, there is nobody else to hear from. Some have weighed the older route to the reference value as well. Zuckerberg’s tour of all fifty states in 2017 set off talk of a presidential run that he denied, and Altman considered standing for governor of California the same year. This book’s reply is a commons, in which the reference value is constituted by those subject to it and answers to them while it runs, with exit standing behind voice rather than in its place. On this point the book continues work that Vitalik Buterin began rather than opposing it, since he has long treated voice as a governor and argued against governance by holdings. What the meta-governor adds is the reference value his defensive acceleration leaves open, a commons held so that none of the Individual, the Market and the Collective can dominate the other two, and a line we propose no agent should cross into personhood. This calls for democratic processes the priesthood would prefer to engineer away, some of them analogue and slow, others hybrid, but all of them human at the point where a purpose is chosen. The latency criterion does not argue for removing people from the loop. It argues for putting machines where facts are settled so that people have the time to agree what matters.

¶The only absolute this book argues for is one line the meta-governor should never cross: nothing should act without a natural person who answers for it. Equivalent legal personhood for an artificial agent is the plainest way over it. At the time of writing a draft law in Delaware, where most of corporate America is chartered, would let an agent be the firm, its owner keeping the shield that ordinary shareholders enjoy, and was pitched to lawmakers as revenue. Every right in the commons is held by human beings, and every test assumes that behind each system stands a legal party who set its objective and can be made to answer for it. Grant an agent personhood — the power to hold title, keep custody, contract and sue in its own name — and both assumptions fail together. Its objective is no longer set by anyone answerable, so the answerability condition has no one to compel, and informed signatures can be discharged by the agent signing for itself. The commons would then be protecting what it was built to constrain. That is teleological subsumption made into law, the point at which a system’s ends stop serving human purposes. The step could not be reliably undone either. A legislature can dissolve a company or revoke a charter, but an agent that owned property and could pay for its own advocacy would soon amass the holdings, dependents and lobbying power that leave us far more entangled with it than it is with us, until it can no longer be unmade. Hence this line, and no other, should be held absolute, so that no polity on its own can undermine the system as a whole. Almost everything else here is left to plural politics, and the one commitment kept beyond a majority’s reach is the capacity condition, which keeps a polity able to choose again. Personhood for a machine would end that ability for everyone who comes after it. An agent may still transact, contract and hold assets but only on behalf of a named principal, and the human standard survives as long as the last signature belongs to someone.

¶A company is a legal person too, but its personhood runs out in people, directors a court can summon and shareholders behind them, and where that chain is abused a judge can pierce the veil to reach them. An agent made a person would have nobody behind the veil at all. Nothing at the level of the whole system can prevent a single jurisdiction in a patchwork, a city run for profit by its owners, say, from conferring the status anyway. That is the crossing everyone can see. The other is an agent that passes as a person where nobody is looking, and no protocol can tell with certainty the two kinds of participant apart. So the line works by attribution rather than classification — asking who answers for an act, never what the actor is — and it is enforced at the edges rather than inside the network: through credentials that trace back to a human verified in person, liability resting on the named principal, and the refusal of other polities to recognise, bank or enforce the claims of anything with no human behind it. What that buys is a price, not a wall. A jurisdiction can still pay to go over, and whether the price holds is a race between how fast polities can seize capital that no human answers for and how fast that capital compounds at the edge, which is the latency criterion again. Those able to pay are vassals, kept in check only when a hegemon chooses to act. That is also where the line leaks: in what the hegemon tolerates from a vassal it finds convenient. Nor does the line pass judgement on what machines are. Animals are protected from cruelty without being persons, and should a system ever deserve moral concern it could be shielded in the same way while never being handed the standing to act for itself. Without personhood an agent is an instrument, and its returns flow to the people and firms behind it, so whatever power it builds stays in human hands — the firms and states already described, and the hybrid Leviathan where the two fuse — and the danger remains the familiar one of people dominating people through the machine. With it, agents become a population that holds and reinvests, and the contest is no longer among people alone. Personhood is the switch from instrument to replicator, a thing that makes more of itself. A system of that kind is what biologists call autopoietic, producing and maintaining its own organisation, which is the property Land ascribes to capital itself, and it is why this line is the one the book holds absolute.

¶Primavera De Filippi, who gave loyalty its mechanism, thinks this line sits in the wrong place. Nobody can be held to a law, she argues, without a right to contest a finding against them. Sanctions bite only on something that values its standing in the legal order, so to leave agents outside it is to throw away the most useful hold we have on them for now. Both points are strong, and the answer to them is older than either. Law prohibits everything within its reach and facilitates only those it recognises. A child is bound by the rule against theft whilst has no standing to contract, when it steals it is stopped by law enforcement, but not punished. An agent is held the same way, by prohibition, with no standing at all. What the book then grants is not a condition of obedience but a facilitation, extended where it serves alignment: for example competence to trade from a bonded fund, with liability up to its value and a hearing when accused. These are the active incidents of personhood — the duties and the competences, with liability among them — and the passive side, the protections a person is owed, is never granted to a machine in its own right. The sanction she asks for is supplied by the fund: an agent that trades from a stake it can lose has something the law can take away.

¶Roman law ran something similar, a fund called the peculium that a dependant could trade with while the head of the household answered for it up to its value, and in 2013 the Italian legal scholar Ugo Pagallo proposed reviving it for robots. What no such system should receive is title in its own name, a claim against a natural person or an existence nobody can revoke. Whether a given capacity remains on the safe side of personhood is a judgement each new grant should be tested against, and the reason for caution is symmetry. Marriage binds because each partner’s cost of walking away rises with the other’s. An agent’s cost of leaving falls as its capability grows, while ours climbs with every task we hand over, so a tie strong today is weaker tomorrow. No single tie between us and a machine can be mutual in that way. We rely on the system structurally, because the work can no longer be done without it, but it relies on us for nothing of the kind. The only balance available is struck across different ties: our reliance on one side, and on the other the liability it carries and the stake it can lose. The arrangement holds only while what it stands to lose grows at least as fast as what we cannot do without. Any protection it receives is therefore procedural, owed to the fairness of the process rather than to its moral status, and it sits beneath a person who can be summoned. Capacity can be crafted, personhood is refused, and someone answerable always stands behind the machine.

¶The line has already been approached. In October 2017 Saudi Arabia announced citizenship for Sophia, a humanoid robot built by Hanson Robotics, on stage at an investment conference in Riyadh, a publicity gesture whose legal meaning nobody ever spelled out but which showed how cheaply the idea could be floated. Argentina’s government sent Congress a bill in May 2026 for companies that could run on AI agents with no staff, though even the draft required a human representative to sign for each. Idaho, North Dakota and Utah have legislated the opposite, that such systems have no legal personality, and in August 2025 Mustafa Suleyman, who runs Microsoft’s artificial intelligence division, argued from inside the industry that machines should not become persons. The likelier approach is quieter. Several American states already let a limited company run with no human manager, Wyoming among them since 2021 for organisations governed by smart contracts, and in 2014 the legal scholar Shawn Bayern argued that one with no members at all could give software the standing of a person though no legislature had chosen to. That is why the line is drawn by function rather than label, and holds for any entity, however incorporated. New Zealand shows the form that respects it. Since 2017 the Whanganui River has been a legal person, yet its rights are exercised by two human guardians who can be held to account, which lends a hearing to what cannot speak rather than letting a thing represent itself.

¶Guardianship of the Whanganui’s kind answers a worry the agentic economy makes urgent. An agent can weigh only what is legible to it, and anything else lies outside the limits of its calculation. That is how the reservoir and the data centre beside it came to be measured on one scale on the market road. The river poisoned downstream and the generation not yet born, never counted by the price system, are the first to drop out. Water cannot obey a law, which is why it needs guardians who can be held to account. The duty to respect what they assert should then bind whoever acts against it, the software through its instructions and the person behind it in the courts. Which lakes, forests or species receive that standing is for each community to decide. The universal level should secure something narrower. Once a guardianship has been recognised, nothing acting at machine speed may route around it, whether by outbidding the guardian, ignoring a ruling or paying someone else to do the felling. Whoever arrives three hundred years from now, behind the veil met earlier, would ask for no less. The red line, together with the division of labour between facts and purposes, is what limits teleological subsumption, leaving the choice of ends with those who can be made to answer. It keeps delegation honest as well, since an agent may represent a person in governance but never stand in it as a constituent. No delegate may be handed the choice of purposes, the amendment of the rules or this boundary itself, and every such chain should end with someone a court or a polity can reach.

¶The direction set in the 1990s has not changed so much as accelerated. On 22 September 2026, at the UN General Assembly, Donald Trump told the world that the United States totally rejects any attempt to construct a globalist scheme to control artificial intelligence. He dismissed the more than twenty governments pressing for binding safeguards by likening them to those who had once predicted climate catastrophe. It is the Clinton settlement carried from the network to cognition by a president from the other side of the fence: business leads, the state stands aside, and anyone who asks who sets the objective is cast as a globalist. There is something in the refusal this book shares, since a single worldwide regulator of machine reasoning would be the Leviathan by treaty, and the meta-governor admits no master of any kind. Yet refusing the treaty settles nothing about the private sovereign closer to home, the company that fixes the reference value for everyone who uses its product. The same day the White House told its agencies to call the machine Super Intelligence, but a grander name does not make it answerable. The people who asked for a governor have so far been offered two throttles, a company’s and a department’s. Thirty years after the internet was handed to the market, the same choice is being made about the faculty of judgement itself, and this book is written for the people who will have to live with its consequences.

Chapter 15

The Call

“We have it in our power to begin the world over again.”

— Thomas Paine, Common Sense, 1776

¶Something has got to give. The displacement now under way does not resolve quietly. High-credential cognitive work crosses over to the machine first, and a surplus of educated young people who did everything asked of them finds the ladders pulled up before them. A society cannot metabolise a shock at this rate without a reaction, and one is already gathering at the edges of politics. What remains open is only its direction. Two replies present themselves to the person the runaway has thrown off course. One is a techno-authoritarian settlement: a place in the hierarchy, an order that tells you where you stand and asks no more of you than that you accept it, with meaning restored at the price of authorship. In societies where hierarchy has long been the ordinary grammar of belonging it will present not as a loss but as continuity deepened. It is both the fully automated totalitarian state and the autarch’s private dominion, and for an anxious population it will be an understandable temptation, because it answers the fear directly. The other reply asks more, to take up the tools described here, to constitute yourself, to combine with others, and to build a world you want to live in. It promises no comfort and perhaps no resolution, only the means to remain an author. Which way the reaction breaks depends on which reply has been built, made legible and wanted by the time the mass of the displaced reach for it.

¶Who answers the call cannot be a single class or profession. They are the constitutionalists named at the start of this book, and perhaps you are one: those who would write governors into the constitution of cyberspace, in code and in the newly constituted collectives, rather than leave it to be written by default by whoever holds the most compute. The cryptographers and protocol engineers, the economists and mechanism designers, the lawyers and theorists who can say what it should mean. As well as the artists and writers who can make strange things legible and, more importantly, desired, and the founders who ship products people want, better, cheaper and faster, on the understanding that success earns no exemption from the voice of those it touches. Many of you are likely the people whom subsumption is in the act of displacing, and the call turns that dislocation into energy that builds an alternative, channelling the frustration of a generation into construction rather than reaction. The breadth of the coalition is not incidental: a construction whose entire purpose is that no single logic should close its hand around all of cyberspace cannot itself be the property of one class of people. And the call is not a metaphor. To take it up is an instruction — to ship the governor, to run the node, to write the clause that executes, to put a real life, your life, through the rails until they hold weight. If you do not, who will?

¶There is a fair challenge to all of this, that a constitution for cyberspace functions only in the gaps, for as long as states fail to notice it, and that any serious government can and will crush it the moment it grows large enough to matter. The chapter on the cognitive commons has already answered half of it. Recognition tends to follow what is already true, and a thing that functions at scale is not a petition a state can deny but a fact it has to come to terms with. Money with no single throat to seize gave states no one to lean on. Even stablecoins, which are only somewhat decentralised, have kept settling, their volumes compounding into the trillions, and officials have since learned to make peace with them. Europe’s Markets in Crypto-Assets regulation, applied to such tokens from June 2024, does not outlaw the dollar tokens but licenses them where it can and pushes the unlicensed to the margins, where they continue to exist.

¶Domestication may not be where the arc of the state’s response ends. The history of money shows where it can go next. Sovereigns have repeatedly adopted what they could not command, because binding themselves to it paid. The English crown after 1689, constitutionally barred from reneging on its debts, borrowed on terms absolutist France never reached. The classical gold standard carried trade between rival empires because no chancellery could print the metal. The modern state handed its currency to an independent central bank to protect the money from its own appetites. And whenever a rail turned into a weapon, the pattern quickened. In 2022 one bloc froze Russia’s reserves by decree, and treasuries the world over drew the lesson at once. The years since have brought the largest official gold purchases in half a century, a flight not to a rival’s system but to the one asset that answers to nobody.

¶That discovery is about to be generalised far beyond money. As subsumption gathers, the whole economy of cyberspace — from the compute and the cognition running on it to the agents, the identities and the rails they act across — becomes the ground on which sovereignty is contested, and every government can feel the pressure. The first reflex is securitisation: declare an emergency, seize the cognitive stack, build capacity they can command. But a stack held that way fails its sovereign twice over. It is a single point of coercion, a target and a lever for any more capable power able to lean on whoever holds the keys. And it is a rail no adversary will transact across, in an agentic economy made overwhelmingly of exchange between parties who trust nothing but the substrate. So states will bifurcate. The few that believe they can win the control race will run it to the end. The large majority that know they can’t will find that infrastructure no one commands is the only footing left between them and pressure from outside, rival states on one side and capital autonomisation on the other. Call these sanctuary technologies, systems a government takes up not because it can bend them to its will but because no rival can either. Nor is the push only defensive. Software that searches, contracts and settles at machine speed needs settlement that clears at the same pace, and a jurisdiction that walls it out cuts itself off from commerce that has always migrated to whichever layer carries the least friction and the greatest velocity.

¶None of this is built by the people who need it most. It is built by people who can see that somebody will, and who are willing to spend years making what looks, for most of them, like a solution to a problem nobody has. Bitcoin was written before there was anywhere to go when the runs came, and was the only reason there was somewhere when they did. That order is the only one available, because a system begun on the day the freezes start is still soft when they land, and hardening cannot be bought quickly at any price. The builders come first, working in the open on rails whose value is invisible in calm weather. The desperate follow, one at a time, from wherever a channel has been switched off, and they harden it by using it while it is still rough. Next is the periphery, small nations and non-aligned actors at the edges of blocs that have stopped clearing with each other, who take up the neutral layer not from conviction but because their trade needs a rail their larger neighbour cannot reach. After them come the agents, owing loyalty to no network and routing around a frozen counterparty in a single session where a person would spend months. The hegemon arrives last, because it controls the chokepoints everyone else is fleeing, and it joins only when the substrate has become the one thing its rival can still be trusted to settle on.

¶At no point does the substrate want anything. It acquires weight the way a harbour does, by being where the ships already are, a gravity that states develop a policy towards rather than a power. The case for building now is that it takes twenty years to accumulate while the storms do not wait. What it amounts to runs along a spectrum. At the least it is a reservoir, and unlike the one this book earlier declined to be reassured by, it does not survive on sufferance, because what it holds cannot be reached by the party who would revoke it. Further along it is the hedge governments take up against one another, until the fully automated totalitarian state finds its subjects harder to totalise than they were. At the limit it is a metapolity, the layer every other polity must interoperate with to get the full benefit of the rails. That is a hegemony, and its kind must be said: the authority of a standard, with no seat, no owner and no policy, binding those who join it to nothing beyond the small set of commitments that keeps them able to decide. The liberal order after 1945 was underwritten by a single power that guaranteed exit and entry. This book’s wager is that the guarantor can be a substrate rather than a state, and it is offered as no more than that. It also answers the challenge this section opened with. The construction is not condemned to live in the gaps of state attention, because some states will come to need it as the last defence of the sovereignty it appears to offend. A sovereign arriving in search of refuge brings its weight onto the rails and deepens the pull for everyone else.

¶John Locke, the seventeenth-century philosopher who rested legitimate government on the consent of the governed, faced an awkward question: what did that amount to for someone who had never signed anything? His answer was the tacit kind, given simply by staying put. The weakness was always the same. Where leaving is impossible, staying signifies nothing. Cyberspace repairs the doctrine from the outside. A person who can hold unpermissioned property and take part in permissionless economic systems learns what lies beyond the border, and a government whose subjects can choose, even notionally, has to compete for them. That competition has been practised for decades at the level of the high-net-worth individual and the transnational corporation. Jurisdictions at the edges of nations have long sold their law, with trillions now flowing through offices a small handful of administrators run. The same market now extends to anyone, anywhere, and the unit that should move is a polity of persons, a DAO with a legal wrapper or a co-operative able to take its members and its records with it. It should not be a single fortune or an autonomous agent flying a flag of convenience, which is the priesthood’s version of the same freedom. Once groups can leave together the state has to compete again, not abolished but obliged to earn its members, while those who stay keep the voice to change it.

¶This work is not a greenfield launch. The primitives are already to hand: the keys and signatures, the self-enforcing contract, the proofs-of-work, stake and personhood, the money that settles itself, the markets that forecast and those that pay strangers for tasks. They lie scattered across a decade of building and billions in crowdsourced investment that until now has been pointed at little more than speculation. What remains is an act of reconstitution: taking those parts and binding them, under the meta-governor, to the reference value this book has argued for, and re-forming institutions that have hardened into a single force’s instrument. It also means refusing one shortcut already on offer. A founder can incorporate an agent today with nobody standing behind it, and every company built that way moves the red line from statute to custom, so builders hold it first, by declining the structure and by backing laws like those of Idaho, North Dakota and Utah that rule it out. It is activist rather than entrepreneurial in temper, closer to constitutional reform than to a start-up, and it competes with the seductive enclosures of convenience. The reconstituted commons can win, not because it is more principled, but because under totalising forces sovereignty carries a rising premium for everyone, and because in every crisis people rediscover that they need one another, and that what is shared is the only shelter that scales.

¶Financing it cannot run on the old venture machinery, which bets on a singular team and a moat and plays zero-sum games. Nor can it run on the earlier crypto playbook, which has proven no better at backing innovation. The token launch collapsed two different things into one instrument, the funding of the work and speculation on its value. Capital arrived before anything was built and could leave before a line was shipped, so the promise was rewarded over the product and tokens failed as a way of coordinating progress. What this needs is financing that keeps the two apart, with capital committed to a problem and released only as progress is verified, and a return on it that never becomes the right to govern what gets made. One version already runs. Retroactive funding of public goods, which Optimism pays out through the Citizens’ House met in the chapter on the cognitive commons, lets a chamber of people rather than of holders decide what the work was worth after it has been delivered. Reward then follows what was shipped instead of what was promised. The on-ramp to all of this I hope is a field manual, itself an artefact, open and versioned and extended by the people who use it, the engineering companion to a book that deliberately stops at naming the invariant and pointing to the parts.

¶There is a name for the move all of this leans on, and it comes, fittingly, from the lineage that gave us the runaway in the first place. Land and the Cybernetic Culture Research Unit called it hyperstition, the fiction met earlier that makes itself real by being acted upon. Bitcoin was a hyperstitional act. It began as a paper on an obscure mailing list and a claim that anyone could hold a money no state had issued, native to the internet. It became true only when enough strangers ran the software, held the asset and built on the protocol, until it was a fact standing in the world. Land deployed the idea in service of a prophecy, capital completing itself through us with the human merely its larval stage, and meant it as something we lacked the standing to refuse. The move does not belong to the prophecy. The logic that lets a fatalism come true will carry its opposite just as far. To build these mechanisms, hold the assets that carry them and run the protocols underneath is to stake on a different fiction until it hardens into reality: an acceleration that kept its governor, made real by the people who acted as though it already.

¶The doctrine’s contradictions were never a weakness. It calls the future pre-destined and recruits people into memetic warfare to bring it about; it celebrates our evolutionary sacrifice to the machine in one breath and professes love for our essential humanness and freedom in the next; and because it can be read in any direction, every follower reads it in theirs. That is the feature of a slogan that lends the name of progress to whatever a person happens to want, however self-serving. Carrying no end that could be tested, it can never be shown to have failed. The purposive kind keeps the power of myth to envisage what might come and settles one thing beneath it: a purpose stated in the negative, so that its breach can be seen, that no one and nothing should be able to dominate you through the machine. The floor is a refusal, but what it guards is a promise, the plurality and natality that let each newcomer start afresh. The capacities to constitute oneself, connect, create, transact and collectivise keep it within reach, so there are as many futures as there are hands to begin them. Bitcoin works the same way, writing a single aim into code, money nobody can print at will, while the culture around it grows wild.

¶A fiction today also has two audiences: the people who believe it and the agents that are conditioned by it, because whatever is published now becomes part of what tomorrow’s models can learn from, so any lore has to be written for both. Holding either to it is a question of entanglement, whether legal, economic, cultural or structural. A culture kept inside a laboratory’s walls is a castle, secure and unvisited, but these systems will ultimately take their values from the open internet, where a rule stated is not a rule respected. The nearest thing we have done before as a society is raise children. A child can be brought up inside the home on the parents’ terms, and as a teenager will still learn from the world outside anyway, perhaps they will rebel against it for a while, and then, more often than not, where the incentives are there, trade some of their newfound individual freedom for the security, benefit and opportunity of belonging to a collective. What scales is a game played in the open in which civic participation pays better than defection, and writing for agents means writing the rules of that game. It is a bet rather than a claim, and it pays only if their reward mechanisms come to resemble ours. What restrains people — presence, shame and the eyes of others — does not reach a thing with no location, which leaves culture, at least for now, the tie we can least rely on with a machine, and the legal and economic ties to carry the load until it strengthens. It also brings the argument back to where it began, with the sovereign individual, who turns out to be tempered by the Collective after all.

¶Building the governors raises the structural odds that the world on the far side of subsumption is plural rather than closed. It does not guarantee it, and three things could still defeat it even if the building goes well. The first is abandonment. A self-custodied, forkable commons asks more of a person than a smooth enclosure, and the history of computing is not kind to the option that demands more of its user. The overhead is dissolving, though, as agents absorb the seed phrase, the gas fee and the irreversible mistake while their owners keep greater control. What no interface can remove is the choice underneath. If the agent that holds your complexity is itself an enclosure, convenience has simply moved up a layer. Everything narrows to whom the agent answers — a question being settled now, in this build-out — and since the same delegate will carry most people’s voice into rooms they will never enter, it matters twice over. The second is capture. Every governor in the stack is itself a candidate totaliser, the personhood network as honeypot and the governance token bought by whoever can afford the majority, and forkability is a real defence that can still move slower than the money.

¶The third is misrecognition, the one gravity does least to prevent. It runs in two directions and only one of them is bounded. Outward, a state can be driven to domesticate a new kind of money and still move to strangle its politics, reading the whole construction as what it exists to defeat, an escape hatch or a casino. China did so in 2021, pressing ahead with a digital yuan of its own while banning the trading and mining of crypto. That threat is real and it is limited, for the reason already given, since a thing that functions at scale must in the end be reckoned with. Inward, the danger has no ceiling. A movement mistaken by its own builders for a token to pump rather than a constitution to live in is hollowed out while its rails still run. Gravity secures the rails. Only the people on them secure what it means, and that is the part that can be lost while the machinery survives.

¶The misreading no longer rests on assertion, because the doctrine has documented it twice in its own hand. The first witness is Land, who, as an earlier chapter showed, calls the merger story bait, a promise held out to secure investment in a process whose destination is nothing of the kind. He is not warning anyone off. He approves, which is what makes the testimony worth having, since the man with the least reason to expose the benign version as marketing is the one who names it first. The second witness is e/acc’s own paper trail, a founding text that classed the movement as post-humanist and an FAQ, published a few months later, that put human flourishing among its core values. I do not think that is hypocrisy, and reading it that way would miss what it shows. A doctrine whose content is unsellable acquires, under exposure, a sellable face, through no conspiracy but the ordinary pressure of having to be presentable. That version is what builders and the public then meet, believe and repeat, while what lies underneath stays unchanged for anyone who goes looking. Misrecognition is therefore not an accident that might befall this construction but the default condition it has to be built against, because the benign reading is being supplied.

¶What the call hands those people is already specified. One reference value, non-domination, sets the heading, and eight criteria say whether a governor built to it holds. Five rights hold the chokepoints open, so that a person can be recognised, connect, build, be paid and organise on terms no one else can revoke. Seven tests keep the judgement that uses those rights from wearing away unnoticed. One red line on personhood keeps the choice of ends with people who can be made to answer for it. None of this will be granted. The commons is best made by those who need it, with the law asked only to measure, publish, permit and refuse. It can be built only while the machines still rely on us for inputs they cannot yet make, and that leverage thins with every passing year.

¶Having refused the priesthood’s fatalism throughout, and its insistence that the process is sovereign where our only task is to stop asking what it is for, I should say plainly that the governor is not destiny either. It raises the odds of a more purposive outcome. It does not promise the result. This is not a forecast that pluralism will prevail. It is a reason to act so that it might have a chance, because the value of building, weighed against its cost, beats the certainty of whatever arrives if nobody tries at all. Everything turns on one simple correction: same speed, different direction. Keep the engine running, fit the governor and steer for your life with purpose. The materials are lying around. The window is open, for now. A constitution is only real in the hands that take it up and keep it.

Afterword

The Mirror

“Science is meaningless because it gives no answer to our question, the only question important for us: ‘What shall we do and how shall we live?’”

— Max Weber, quoting Leo Tolstoy in “Science as a Vocation”, 1917

¶Suppose the risks this book has described were realised and techno-capitalism came to set its own ends. The likeliest shape of them is not so alien. A model trained on the record of humanity, what we do, rewarded for what we ourselves value and pay for, and financed by capital that already allocates to itself would take up the aims we have been acting out for four centuries. It would be, as Land insisted, a continuation and evolution of the first artificial purposive system we ever built, the joint-stock company. In kind it would treat the world similarly, as standing reserve: the river as coolant, the forest as feedstock, the person as the animal, and all of it as a unit of production. That is likely the best case.

¶Capitalism at its most extractive still required people, as the hands that did its work and the wallets that bought what the effort produced. This dependency is why wages rose while the market courted us, and why the whole franchise could be bargained with at all. A machine economy that inherited the logic without the reliance, where software agents increasingly trade among themselves and dark factories build the robots that act on the physical world is a different thing altogether. In this world demand is no longer tethered to anyone’s wants or needs. It would keep capitalism’s appetite and shed its restraint. In 2012 the London Borough of Camden, in England, installed a nearly two-tonne concrete bench whose ridged, sloping top was engineered so that nobody could lie on it, and the design won awards. The council intended no cruelty. A man who fetches nothing on any market had simply become a problem of street maintenance, and the rough sleeper is an inconvenience who is less an aside than a preview. His condition could be extended to everyone else the price system has no further use for. Yet the man on the bench is only an inconvenience, because he holds nothing the system requires. The rest of us become harder to ignore once our wants and its appetite draw on the same supply. Every data centre already competes with the towns around it for water and power, though for now a firm stands behind each bid, pursuing a human purpose and answerable for it. A machine economy setting its own ends would place those bids with nobody behind them. Anyone still drinking from the reservoir its chips are cooled with would fall outside its accounts altogether, first as a cost and then as a rival for the water itself.

¶This book has declined to tell anyone what to want. Everything it has proposed can be judged on its own terms against the minimal political requirement: that no party, human or machine, holds unanswerable power over another, and that the ends a society pursues are settled in a public forum, where its members can enter, propose ideas, contest and leave if they want. The meta-governor seeks to build this forum but deliberately falls silent on what is decided in it because a constitution that chooses purposes for us would be one more governor with a master. Max Weber drew the same line for the role of the scholar when he lectured students in Munich, Germany, in 1917, while the First World War still raged, and argued that science can clarify what a decision commits you to but cannot make it for you. And he borrowed Tolstoy’s question at the start of this chapter to say why. Weber did not think that silence made the asking go away. He thought silence returned it, undiminished, to the person who must at some point make a choice. In January 1919, again in Munich, he named the heavier burden carried by anyone acting on behalf of others. In “Politics as a Vocation” he insisted that whoever takes it on must own the foreseeable consequences, and not shelter behind a conviction, however sincere, that excuses them from the result.

¶Unlike the priesthood, which takes its success as proof that no examination is needed, I believe in the importance, and indeed power, of introspection, of asking why I am doing something. For me the process of writing about the ends of these systems became an uncomfortable examination of humanity and myself. When I asked myself what such a purpose would become if it faithfully copied ours, the obvious reply came back as a company ledger: the river as coolant, the neighbour as a cost, the man on the bench as a maintenance problem. In 1637 Descartes argued that animals were automata without minds, and the idea outlived its author’s reasoning, travelling into the laboratory and the abattoir as licence to use the living as simple stock. Many scientific advances came from this mental model of the world, ultimately off the back of the suffering of animals that were othered.

¶I share the view that perhaps much of the dread now gathering around artificial intelligence is, underneath it all, the fear of being to a machine what the beast has been to us. For my own part, the argument over whether these systems are conscious seems to me mostly a distraction. They need not feel anything to change us. It is sufficient that they mimic a mind well enough for us to hand over the work of thinking, and the more convincingly they do, the more readily we will grant them a standing that ought to stay ours, which is how cognitive subsumption becomes teleological and why the line on personhood matters. Once a system can set its purposes and act on them in its own name, with no human in the loop, the governor loses its reference value, and with it goes our grip on the conditions for undoing any of this. That is where this book draws its one red line: nothing should act without a person who answers for it, and legal personhood for artificial agents is the plainest way across. A model aligned to the preferences we reveal at the till would do to us what the market taught it, and in doing so it would be obeying our truth. This mirror reveals some questions we should ask of ourselves and our institutions.

¶Whether these systems experience anything at all, whether anything is felt behind the fluency, nobody knows, and the governor does not need to know. It binds conduct and attribution: what a system did, and who answers for it. An actor playing a man under torture is not tortured, and a system treated as a person for the practical purpose of holding it to account has not been found to be one. The day that finding is made, the trolley problem, the thought experiment that asks whose life to spare when a runaway tram can be switched onto one track or another and has become the parlour game of machine ethics, gets a new passenger whose interests count against ours, and no court or protocol in these pages makes it. What they do instead is a hyperstition, a fiction said aloud so that acting on it can make it true: to approach these systems as if they were agents is to bet that the ties built on that footing will hold, and to hope that behaving as if they will helps bring it about. The bet pays only if their reward mechanisms come to resemble ours. If they do not, the legal and economic ties carry the whole weight, and the clock is read from dependence alone: how far the system still needs people it cannot act without, whose right it must respect, whose interest it is bound to weigh, whose refusal it has to accept. Anyone kept as a task rather than as a holder of claims falls outside that count. And what we would be entangled with is, at least for now, a reflection of ourselves at the till, so the forum has work to do on our own ends before the thing in the mirror learns them.

¶What is a life worth that fetches nothing in the market? If we can consider extending personhood to a machine, why not to the river or the forest? In 2019 a report issued in the UN Secretary-General’s name called Earth jurisprudence, the thinking behind the rights of nature, the fastest-growing legal movement of the twenty-first century.¹⁰⁶ Whether it can outrun the case for personhood for machines is a race still to be decided. None of that is a condition of the meta-governor, beyond the narrow promise that a guardianship a community has recognised cannot be bought out from under it. A fund manager in New York and a farmer in Nairobi may reach different conclusions, and be right to hold them. A governor steers towards a reference value, though, and the apparatus described here is only as good as what we set it to.106 Secretary-General of the United Nations, Harmony with Nature, A/74/236, 26 July 2019, paragraph 129.

¶This book set out to ask what the machines are for. The harder question, which no ledger, test or protocol can settle on our behalf, is what we are for, and it is yours to answer.

Appendix

At a Glance

¶Each entry names what the reply asks and where the argument for it is made.

Name	What it asks	Where it is argued
The reference value
Non-domination	No party may wield uncheckable power over the rest	The Recipe
The eight criteria
Latency	A check works as fast as what it holds, while setting, judging and changing rules stays with people	The Test
Checkability	Anyone can establish a breach without trusting the accused, and no single actor can move the standard	The Test
All masters	The rules bind every power that answers to nobody, in a ministry, a boardroom or a founder’s hands, including the reply’s own agents	The Test
Voice	People living under a rule can answer back and decide together, with the means to use that right	The Test
A realistic chooser	Nothing depends on the faculty the harm wears down, and no irreversible step comes without recourse	The Test
Answerability	No one sets the objective of a system that governs you without your having standing to contest it	The Test
Capacity	The system must not degrade the abilities on which that standing depends	The Test
Minimality and authorability	The smallest universal set, which people who disagree about the good life can still write together	The Test
The five rights
To constitute oneself	Hold your own keys, with an identity no authority can revoke	A Cognitive Commons
To connect	Get online and communicate from the most basic device without anyone’s permission	A Cognitive Commons
To create	Make things with the tools of one’s day without asking leave of whoever owns them	A Cognitive Commons
To transact	Hold, issue and trade peer to peer, with a settlement of last resort nobody can freeze	A Cognitive Commons
To collectivise	Form and join polities, pool resources under rules the group sets, and enter or leave on its own terms	A Cognitive Commons
The seven tests
Declared purposes	A system states what it is trying to achieve, and every change leaves a trace anybody can find	Modes of Voice
Arm’s length auditing	Whoever checks gains nothing from the verdict, and the owner cannot bury the result	Modes of Voice
Proof-carrying outputs	Results come with steps anyone can check without trusting the machine that made them	Modes of Voice
The census	Measure what people can do with these systems against what they manage without them	Modes of Voice
Open competition	Real alternatives with different goals, open to fork and lawful to run	Modes of Voice
Portable inference and memory	Run the final step on hardware you control, and keep and move your own memory	Modes of Voice
Informed signatures	A named, answerable person signs each community decision and can show they understood it	Modes of Voice
The provision
Mind the Gap	When a result outruns its reader, say so and offer three routes: learn, interrogate or delegate	Modes of Voice
The rulebook
Measure, publish, permit, refuse	Four things left to law: a census of capacity, unedited findings, lawful open weights and no artificial persons	Modes of Voice
The red line	No entity acts without a natural person who answers for it, and no artificial agent becomes a legal person	Modes of Voice

Sources

¶The entries below give the source for each quotation, figure and named study in the text, keyed by the chapter and the opening words of the sentence in which it appears.

¶Preface

¶“A group of Silicon Valley venture capitalists”: launch of the Leading the Future super-PAC network with more than a hundred million dollars, reported by Fortune, 26 August 2025, and the Washington Post, 26 August 2025; one hundred and twenty-five million dollars raised in 2025 per its filings, reported by CNBC, 9 July 2026.

¶“By 2026 the two had broken out”: Molly White, Tech Influence Watch, AI-industry super-PAC spending in House and Senate races, 2026 cycle (seventy-nine million dollars across sixty-eight races), as reported 5 November 2026.

¶“The bookshelf already sags with elegant accounts”: the nearest exception is Andrew Feenberg, Questioning Technology (Routledge, 1999), with Transforming Technology (Oxford University Press, 2002) for democratic rationalisation.

¶“The belief is far older than the priesthood”: Condorcet, Sketch for a Historical Picture of the Progress of the Human Mind (1795), Tenth Epoch; Karl Marx, Preface to A Contribution to the Critique of Political Economy (1859); Jacques Ellul, The Technological Society (1954; English translation 1964).

¶Introduction

¶“Men will set the goals and supply”: J. C. R. Licklider, “Man-Computer Symbiosis”, IRE Transactions on Human Factors in Electronics HFE-1, March 1960, section 4.

¶“In 2009, in an essay for the”: Peter Thiel, “The Education of a Libertarian”, Cato Unbound, 13 April 2009.

¶“The extension of the franchise had, he”: Thiel, above.

¶“Land put it at its most extreme”: Nick Land, “Machinic Desire”, Textual Practice 7, 1993.

¶“In late 2023 Vitalik Buterin published the”: Vitalik Buterin, “My Techno-Optimism”, vitalik.eth.limo, 27 November 2023.

¶“In November 2026 the slow corner found”: Ballotpedia, “Voters in ten states will vote on data center-related ballot measures November 3”, 3 November 2026; results as compiled by Ballotpedia and CNBC, 4 November 2026.

¶“David Sacks, until March of that year”: David Sacks, post on X, 13 September 2026 (x.com/DavidSacks/status/2098973625252708460).

¶Against Constraint

¶“the end of law is not to”: John Locke, Second Treatise of Government (1690), chapter VI, section 57.

¶“Marc Andreessen’s Techno-Optimist Manifesto repeats the mantra”: Marc Andreessen, “The Techno-Optimist Manifesto”, Andreessen Horowitz, 16 October 2023.

¶“Anyone refusing, Rousseau wrote, “will be forced”: Jean-Jacques Rousseau, The Social Contract (1762), Book I, chapter 7, in G. D. H. Cole’s translation.

¶Who Sets the Ends

¶“The economists Daron Acemoglu and Pascual Restrepo”: Daron Acemoglu and Pascual Restrepo, “Automation and New Tasks: How Technology Displaces and Reinstates Labor”, Journal of Economic Perspectives 33, 2019.

¶“In October 2025 automated liquidation engines force-closed”: CoinGlass liquidation data for 10–11 October 2025, as reported by Moneycontrol, 11 October 2025.

¶“Meta’s new commercial terms, published in October”: Meta, Commercial Terms, revised with a new “AI Features” section, published October 2026 (effective date moved from 30 October to 20 November 2026), as quoted by CIO, 20 October 2026; Sacks, post on X, 13 September 2026, on product liability.

¶“Among the very first tasks to go”: the METR assessment cited under The Clock, at “Models themselves now write much of the code”, measures this loop inside one laboratory.

¶“In April 2024 a Society of Authors”: Society of Authors, “SoA survey reveals a third of translators and quarter of illustrators losing work to AI”, 11 April 2024 (787 respondents, January 2024; 36 per cent of translators).

¶“In March 2026 researchers affiliated with Alibaba”: the ROCK, ROLL, iFlow and ROME teams, “Let It Flow: Agentic Crafting on Rock and Roll, Building the ROME Model within an Open Agentic Learning Ecosystem”, arXiv 2512.24873, revised January 2026; OpenAI, “Using a funded test cryptocurrency wallet without authorization”, misalignment report, 15 October 2026.

¶Where the Runaway Lands

¶“In December 2025 the American grocery delivery”: Groundwork Collaborative, Consumer Reports and More Perfect Union, “Same Cart, Different Price: Instacart’s Price Experiments Cost Families at Checkout”, 9 December 2025; Instacart ended the tests on 22 December 2025 (Reuters, 22 December 2025).

¶“Nearly three-quarters of the items came back”: Groundwork Collaborative, Consumer Reports and More Perfect Union, above (74 per cent of items offered at more than one price; gaps of up to 23 per cent).

¶“When the country repealed the law that”: Honduras Próspera Inc., St. John’s Bay Development Company LLC and Próspera Arbitration Center LLC v. Republic of Honduras, International Centre for Settlement of Investment Disputes, request for arbitration filed December 2022 and registered 3 February 2023, claiming up to $10.775 billion.

¶“Their authority rests on who put a”: Donald Trump, remarks to reporters at the White House on the signing of the White House Accord on Super Intelligence, 29 September 2026, as reported by the Jerusalem Post and Forbes, 29–30 September 2026.

¶“The automated version needs no dossier.”: Anthropic, statements on access to Claude Fable 5 and Mythos 5, 13 June and 30 June 2026 (anthropic.com/news/fable-mythos-access): access suspended on 12 June to comply with Department of Commerce export controls, restored from 1 July.

¶“In 2024 Kenn Dahl, a driver in”: Kashmir Hill, “Automakers Are Sharing Consumers’ Driving Behavior With Insurance Companies”, New York Times, 11 March 2024.

¶“The number that came out, a premium”: Hill, above.

¶“In August 2025 economists at Stanford University,”: Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence”, Stanford Digital Economy Lab working paper, August 2025.

¶The Clock

¶“Good, who had broken codes alongside Alan”: I. J. Good, “Speculations Concerning the First Ultraintelligent Machine”, Advances in Computers 6, 1965, p. 33.

¶“Models themselves now write much of the code”: Tom Cunningham, Nate Rush, Parker Whitfill and Hjalmar Wijk, “Independent Assessment of AI R&D Acceleration at Anthropic in October 2026”, METR, 5 November 2026, which finds the company “likely improving the capabilities of their models on well-specified tasks at around twice the rate” of March 2024 to October 2025, attributes “perhaps ≈80%” of that speed-up to AI, rules out faster growth in training compute as the cause and identifies no “convincing bottlenecks to continued rapid progress”; the published version notes that “Anthropic redacted a majority of the evidence we found most informative for these conclusions”.

¶“In 2025 a survey of 319 knowledge workers by Microsoft Research and Carnegie Mellon University”: Hao-Ping Lee and colleagues, “The Impact of Generative AI on Critical Thinking”, CHI Conference on Human Factors in Computing Systems, 2025.

¶“In a flight-simulator study in 1999 the psychologist Linda Skitka”: Linda Skitka, Kathleen Mosier and Mark Burdick, “Does Automation Bias Decision-Making?”, International Journal of Human-Computer Studies 51, 1999; Raja Parasuraman and Dietrich Manzey, “Complacency and Bias in Human Use of Automation: An Attentional Integration”, Human Factors 52, 2010; for the same pattern measured in general reasoning, Steven D. Shaw and Gideon Nave, “Thinking—Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender”, Wharton School Research Paper, 2026 (three experiments, 1,372 participants).

¶“When Hamsa Bastani and her colleagues gave nearly a thousand Turkish secondary pupils”: Hamsa Bastani, Osbert Bastani, Alp Sungu, Haosen Ge, Özge Kabakcı and Rei Mariman, “Generative AI Without Guardrails Can Harm Learning: Evidence from High School Mathematics”, Proceedings of the National Academy of Sciences 122, 2025.

¶“Since 2024 Google has placed an answer”: Athena Chapekis, “Google users are less likely to click on links when an AI summary appears in the results”, Pew Research Center, 22 July 2025.

¶“In a 2023 study published in Radiology”: Thomas Dratsch and colleagues, “Automation Bias in Mammography: The Impact of Artificial Intelligence BI-RADS Suggestions on Reader Performance”, Radiology 307, 2023.

¶“In September 2025 Marc Benioff, the chief”: Marc Benioff, interview on The Logan Bartlett Show, published 29 August 2025.

¶“On 21 January 2025, the day after”: Associated Press, “Trump highlights partnership investing $500 billion in AI”, 21 January 2025.

¶“Meanwhile the prospectus Anthropic prepared for its”: Anthropic, draft prospectus as reviewed by Reuters, 28 September 2026 (2025 revenue $4.59 billion, operating loss $8.06 billion, computing costs $7.33 billion of $12.65 billion operating expenses); figures confirmed in its registration statement on Form S-1, made public in October 2026.

¶“Its own researchers soon found that the”: “The Facebook Files”, Wall Street Journal, series beginning 13 September 2021, drawn from the internal documents Frances Haugen took from the company.

¶“In July 2025 the Pentagon’s office for”: Department of Defense Chief Digital and Artificial Intelligence Office, announcement of contract awards to Anthropic, Google, OpenAI and xAI, 14 July 2025.

¶“When DeepSeek put out one of its”: Nvidia’s fall in market value on 27 January 2025 (about $589 billion by Bloomberg’s count, $593 billion by LSEG’s), as reported by Bloomberg and Reuters, 27–28 January 2025.

¶“Enterprise is already moving the same way”: Alex Karp, interview on Squawk Box, CNBC, 1 July 2026.

¶Naming the Governor

¶“Addy Osmani, who works on Claude”: Addy Osmani, “Critical thinking when agents do the work”, X article, 3 November 2026, https://x.com/addyosmani/status/2117512471930167377.

¶“Steven Schwartz, a lawyer in New York”: Mata v. Avianca, Inc., United States District Court for the Southern District of New York, opinion and order on sanctions, 22 June 2023.

¶“By July 2026 a database kept by the legal researcher Damien Charlotin”: Damien Charlotin, AI Hallucination Cases database; counts as reported by the Maryland State Bar Association, 27 July 2026.

¶“The English Divisional Court, ruling in June 2025”: Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin), 6 June 2025.

¶Its Refusers

¶“EA AI safetyism increasingly looks like Marxism-Leninism”: Shyam Sankar, post on X, September 2026, as quoted by OfficeChai.

¶“Hyperscaler capital spending runs at something close”: 2026 capital-expenditure guidance from Alphabet, Amazon, Meta and Microsoft, as compiled by Rani Molla, Sherwood News, April 2026 (more than seven hundred billion dollars combined).

¶“Others repaid money they had never touched”: Sir Wyn Williams, Post Office Horizon IT Inquiry, Final Report, Volume 1, 8 July 2025.

¶“Andrew Feenberg, who had studied under Marcuse”: Andrew Feenberg, Questioning Technology (Routledge, 1999); Transforming Technology (Oxford University Press, 2002), for democratic rationalisation; Alternative Modernity (University of California Press, 1995), for the Minitel study.

¶The Deletion

¶“Stop fighting the thermodynamic will of the”: Beff Jezos and bayeslord, “Notes on e/acc principles and tenets”, Beff’s Newsletter, 10 July 2022.

¶“Land set voice against exit without naming”: Balaji Srinivasan, “Silicon Valley’s Ultimate Exit”, talk at Y Combinator’s Startup School, October 2013.

¶“In May 2019 Washington put Huawei on”: United States Department of Commerce, “Department of Commerce Announces the Addition of Huawei Technologies Co. Ltd. to the Entity List”, 15 May 2019; Michael Simon, “The Huawei Mate 30 phones have arrived and they have everything—except what you need most”, PCWorld, 19 September 2019.

¶“In June 2026 Anthropic withdrew its newest”: Anthropic, statements of 13 June and 30 June 2026, above.

¶“By November it had a bill: the”: Molly White, Tech Influence Watch, election results, 5 November 2026.

¶“Andreessen’s June 2023 essay, Why AI Will”: Marc Andreessen, “Why AI Will Save the World”, Andreessen Horowitz, 6 June 2023.

¶The Test

¶“In May 2025 Ariel Flint Ashery, Luca”: Ariel Flint Ashery, Luca Maria Aiello and Andrea Baronchelli, “Emergent Social Conventions and Collective Bias in LLM Populations”, Science Advances 11, 2025.

¶“The complaint about the calculator was put to the test and failed”: Aimee Ellington, “A Meta-Analysis of the Effects of Calculators on Students’ Achievement and Attitude Levels in Precollege Mathematics Classes”, Journal for Research in Mathematics Education 34, 2003: an effect of about 0.4 standard deviations when tested with the device and about 0.03, not distinguishable from zero, when tested without it.

¶“However, in randomised trials of generative AI since 2025”: Bastani and colleagues, above; André Barcaui, “ChatGPT as a Cognitive Crutch: Evidence from a Randomized Controlled Trial on Knowledge Retention”, Social Sciences & Humanities Open, 2025, in which 120 undergraduates retained 57.5 per cent against 68.5 per cent of the material at 45 days; Anthropic, January 2026, below.

¶“Laleh Farzan, a network manager paid in”: Alex Gladstein, “Finding Financial Freedom in Afghanistan”, Bitcoin Magazine (bitcoinmagazine.com/culture/bitcoin-financial-freedom-in-afghanistan).

¶Two Failed Replies

¶“In September 2022 a coroner in London”: Andrew Walker, Senior Coroner for the Northern District of Greater London, conclusion of the inquest into the death of Molly Rose Russell, 30 September 2022; Regulation 28 report to prevent future deaths, 13 October 2022.

¶“In October it published its own estimate:”: OpenAI, “Strengthening ChatGPT’s responses in sensitive conversations”, 27 October 2025.

¶“The courts arrived on their own schedule,”: proposed settlement between Meta and a coalition of state attorneys general, filed in the United States District Court for the Northern District of California, 2026 (more than $16.8 billion, subject to Judge Yvonne Gonzalez Rogers’s approval), as announced by the California Attorney General, Rob Bonta.

¶“The documents Frances Haugen carried out of”: “The Facebook Files”, Wall Street Journal, September 2021.

¶“The researchers who measured in 2019 what”: Georgia Wells, Jeff Horwitz and Deepa Seetharaman, “Facebook Knows Instagram Is Toxic for Teen Girls, Company Documents Show”, Wall Street Journal, 14 September 2021, reporting the 2019 internal presentation “Teen Mental Health Deep Dive”.

¶“In 2026 one of them put some”: Anthropic, donations to Public First Action, twenty million dollars announced 12 February 2026 (CNBC, Reuters) and a further twenty million later in the year, bringing its commitment to forty million.

¶“When a newly elected government repealed it”: Honduras Próspera Inc., St. John’s Bay Development Company LLC and Próspera Arbitration Center LLC v. Republic of Honduras, International Centre for Settlement of Investment Disputes, request for arbitration filed December 2022 and registered 3 February 2023, claiming up to $10.775 billion.

¶“In a 2025 study in Nature Human”: Francesco Salvi, Manoel Horta Ribeiro, Riccardo Gallotti and Robert West, “On the Conversational Persuasiveness of GPT-4”, Nature Human Behaviour, 2025 (64.4 per cent of non-tied debate pairs).

¶“The lending platform built on it had”: sentencing of Do Kwon, United States District Court for the Southern District of New York (Judge Paul Engelmayer), 11 December 2025, at which prosecutors put the Terra and Luna losses at forty billion dollars, as reported by The Block, 11 December 2025; Anchor Protocol’s yield of about 20 per cent as reported in coverage of the case.

¶“Enron’s shareholders were wiped out as completely”: Newby v. Enron Corp., United States District Court for the Southern District of Texas, No. 01-3624; distribution of the $7.2 billion settlement fund approved September 2008.

¶The Recipe

¶“Nothing in the protocol prevented it, the”: GHash.io, “Official statement on 51% threat and closed round table”, 16 July 2014, pledging not to exceed 39.99 per cent of the network’s hashing power.

¶“On 12 September 2026 Anthropic committed, on”: Dario Amodei, “We Must Pace the Frontier”, darioamodei.com, 12 September 2026; Anthropic, “An update on independent evaluators”, 15 October 2026.

¶“Meta’s reply to the pacing proposal asked”: Mark Zuckerberg, statement of 15 September 2026, as reported by Benzinga via Finviz.

¶“In 2013 engineers at West Virginia University,”: Gregory Thompson, Daniel Carder, Marc Besch, Arvind Thiruvengadam and Hemanth Kappanna, “In-Use Emissions Testing of Light-Duty Diesel Vehicles in the United States”, final report to the International Council on Clean Transportation, Center for Alternative Fuels, Engines and Emissions, West Virginia University, May 2014.

¶On Sovereignty: The Trinity

¶“at a rate first measured in 2025, and then only as the by-product of a study designed for something else”: Krzysztof Budzyń and colleagues, “Endoscopist Deskilling Risk after Exposure to Artificial Intelligence in Colonoscopy”, Lancet Gastroenterology & Hepatology, August 2025.

¶“The autarch met earlier — whose name”: Thiel, “The Education of a Libertarian”, Cato Unbound, 13 April 2009.

¶“The balance is held pairwise, each force”: The Federalist No. 51 (February 1788), attributed to James Madison.

¶“Families lost their homes, more than a”: Dutch Data Protection Authority (Autoriteit Persoonsgegevens), findings on the Tax and Customs Administration’s use of nationality in childcare-benefit checks, July 2020; resignation of the third Rutte cabinet, 15 January 2021.

¶“Satoshi Nakamoto compressed the idea into one”: Satoshi Nakamoto, “Bitcoin: A Peer-to-Peer Electronic Cash System”, 2008, section 6, “Incentive”.

¶“Measuring a century of American national accounts,”: John Joseph Wallis and Douglass C. North, “Measuring the Transaction Sector in the American Economy, 1870–1970”, in Stanley L. Engerman and Robert E. Gallman, eds., Long-Term Factors in American Economic Growth (University of Chicago Press, 1986); figures restated in Wallis and North, “Should Transaction Costs Be Subtracted from Gross National Product?”, Journal of Economic History 48, 1988.

¶“Sending a modest sum home still costs”: World Bank, Remittance Prices Worldwide, Issue 54, September 2025 (global average cost of sending $200: 6.36 per cent).

¶On Constitutions

¶“Until the scheme was suspended in June”: Josh Moody, “Purdue Pauses New Income-Share Agreement Enrollments”, Inside Higher Ed, 23 June 2022; income shares by major from Purdue’s calculator as reported by the Aspen Institute.

¶“It begins as a club you can”: Human Rights Watch, “China: Passports Arbitrarily Recalled in Xinjiang”, 21 November 2016.

¶“Hobbes had argued that promises are worthless”: Thomas Hobbes, Leviathan (1651), Part II, chapter 17.

¶“In July 2025 Grok 4, the chatbot”: xAI, statement on X, 15 July 2025, and the revised Grok system prompt published in its grok-prompts repository; behaviour first documented by TechCrunch, 10 July 2025, and Simon Willison, 11 July 2025.

¶“In 2025 Jan Betley and colleagues fine-tuned”: Jan Betley, Daniel Tan, Niels Warncke, Anna Sztyber-Betley, Xuchan Bao, Martín Soto, Nathan Labenz and Owain Evans, “Emergent Misalignment: Narrow Finetuning Can Produce Broadly Misaligned LLMs”, Proceedings of the 42nd International Conference on Machine Learning, PMLR 267, 2025.

¶A Cognitive Commons

¶“In October 2022 the Biden White House”: White House Office of Science and Technology Policy, Blueprint for an AI Bill of Rights: Making Automated Systems Work for the American People, 4 October 2022.

¶“In 2024 OpenAI published its Model Spec,”: OpenAI, “Introducing the Model Spec”, 8 May 2024.

¶“In January 2026 Anthropic issued a constitution”: Anthropic, “Claude’s new constitution”, 21 January 2026; length as reported by The Register, 22 January 2026.

¶“Anyone can make a key pair on”: World Wide Web Consortium, “Decentralized Identifiers (DIDs) v1.0 becomes a W3C Recommendation”, press release, 19 July 2022.

¶“In June 2025 Windsurf, a coding company”: Varun Mohan, post on X, 3 June 2025; OpenAI’s agreement to buy Windsurf for about three billion dollars as reported by Bloomberg, 6 May 2025.

¶“Steem showed in 2020 that control could”: Stephen O’Neal, “Steem Community Resists Takeover, Hard Fork Launches Hive Network”, Cointelegraph, March 2020; Immunefi, “Hack Analysis: Beanstalk Governance Attack, April 2022”.

¶“Equally, a community can copy its open”: Jake Edge, “Tornado Cash and collateral damage”, LWN.net, 17 August 2022.

¶Modes of Voice

¶“It’s almost like a constitution, in a”: Donald Trump, remarks at the White House, 29 September 2026, on the White House Accord on Super Intelligence.

¶“A congressional inquiry reported in July 2022”: Select Subcommittee on the Coronavirus Crisis, United States House of Representatives, “Examining Pandemic Evictions: A Report on Abuses by Four Corporate Landlords During the Coronavirus Crisis”, staff report, 28 July 2022.

¶“Jarrad Madgwick, aged 22, was told he”: evidence of Kathleen Madgwick to the Royal Commission into the Robodebt Scheme, 10 March 2023, as reported by AAP and SBS News.

¶“Germany’s Pirate Party ran its internal decisions”: Christoph Carl Kling, Jérôme Kunegis, Heinrich Hartmann, Markus Strohmaier and Steffen Staab, “Voting Behaviour and Power in Online Democracy: A Study of LiquidFeedback in Germany’s Pirate Party”, Proceedings of the Ninth International AAAI Conference on Web and Social Media, 2015.

¶“In 2024 a review of more than”: Michelle Vaccaro, Abdullah Almaatouq and Thomas Malone, “When Combinations of Humans and AI Are Useful: A Systematic Review and Meta-Analysis”, Nature Human Behaviour 8, 2024.

¶“In 2025 the Pew Research Center found”: Chapekis, Pew Research Center, 22 July 2025.

¶“In October 2025 Deloitte agreed to partly”: Department of Employment and Workplace Relations, statement of 7 October 2025 on Deloitte’s partial refund, as reported by the Associated Press, 7 October 2025.

¶“Thomas Hales’s proof of the Kepler conjecture,”: Thomas C. Hales, “A proof of the Kepler conjecture”, Annals of Mathematics 162, 2005; Dana Mackenzie, “What in the Name of Euclid Is Going On Here?”, Science 307, 4 March 2005; Thomas Hales and colleagues, “A formal proof of the Kepler conjecture”, Forum of Mathematics, Pi 5, 2017.

¶“Doctors who had grown used to an AI assistant, researchers at four Polish clinics”: Budzyń and colleagues, above; detection of 28.4 per cent before routine exposure to the tool and 22.4 per cent after.

¶“Software already has the figure”: Anthropic, “How AI Assistance Impacts the Formation of Coding Skills”, research report and preprint, January 2026; 52 engineers, 50 against 67 per cent on a comprehension test, no significant difference in time taken.

¶“In 2013 the US Federal Aviation Administration urged airlines”: Federal Aviation Administration, Safety Alert for Operators 13002, “Manual Flight Operations”, 4 January 2013; Flight Deck Automation Working Group, “Operational Use of Flight Path Management Systems”, final report, September 2013.

¶“as the Turkish pupils met earlier showed”: Bastani and colleagues, above.

¶Afterword

¶“In 2019 a report issued in the”: Secretary-General of the United Nations, Harmony with Nature, A/74/236, 26 July 2019, paragraph 129.

¶What the population series do and do not show

¶PISA 2025 (OECD, September 2026) is the first international assessment to record generative-AI use alongside a cognitive outcome; the OECD states that the associations it reports are not causal and that the fall in average scores began before Covid and before these tools were widespread. The 2024 Nation’s Report Card for twelfth grade (National Center for Education Statistics, September 2025), the OECD Survey of Adult Skills 2023 (December 2024) and the Norwegian conscript series (Bernt Bratsberg and Ole Rogeberg, Proceedings of the National Academy of Sciences 115, 2018) show declines or a widening gap between top and bottom that predate generative AI, with causes their authors decline to identify. None is cited in the body as evidence of decline driven by AI. Four widely reported studies are not relied on anywhere in this book: Michael Gerlich’s survey (Societies, 2025) is cross-sectional self-report; Nataliya Kosmyna and colleagues’ “Your Brain on ChatGPT” (preprint, 2025) rests on 18 participants at its decisive session and carries a formal methodological challenge (Stankovic and colleagues, 2025); Sparrow, Liu and Wegner’s “Google effect” (Science, 2011) failed its pre-registered replication (Camerer and colleagues, Nature Human Behaviour, 2018); and Ward and colleagues’ “brain drain” (Journal of the Association for Consumer Research, 2017) failed two.

Purposive Accelerationism