--- title: Chapter Three — Where the Runaway Lands book: Purposive Accelerationism author: '@jamie247' position: 7 of 24 canonical: https://whatispacc.org/ebook/#ch3 version: 1.1 language: en-GB licence: all rights reserved; quote with attribution --- # Chapter Three — 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. Nearly three-quarters of the items came back priced differently, by as much as 23 per cent. 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. *** 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. 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”. 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. *** 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. 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. 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: The six relations between the individual, the market and the state. Interactive at the page.] 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. 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. That is an order in everything but name, and it has authors. 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. That is a combustible position in which to put a generation. 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.