Beat China, Become China?
Watch you always, watch us never. The tech and finance elite welcome total surveillance because they've already decided they'll be the watchers, not the watched.
A few years ago, a Chinese journalist named Liu Hu tried to buy a plane ticket and found that he couldn’t.
He had spent years reporting on official corruption, and he was used to the costs of that. This was something else. He went to book a flight and the system simply refused him. It told him he was not qualified. No warrant had been served. No file had been handed over. No one had called to explain. A friend finally pointed him to a court website, where he found his own name on a public list, and next to it the explanation: this person refuses to fulfill his obligations even though he is able to. The planes were closed to him. So were the fast trains, a loan, an apartment. All at once, by a database he had not known existed until it had already decided who he was.
“There was no file, no police warrant, no official advance notification,” he said. “They just cut me off from the things I was once entitled to.” And then: “What’s really scary is there’s nothing you can do about it. You can report to no one. You are stuck in the middle of nowhere.”
Liu Hu was not undone by a vast artificial intelligence handing every citizen a single, sliding score, the Black Mirror version of China that has lodged itself in the Western imagination. That system does not exist. The researchers who actually read the Chinese documents will tell you the real thing is fragmented, barely digitized, and mostly pointed at companies. What put Liu Hu on the list was older and duller than any algorithm. A court had ruled against him; he owed a judgment; and by one careful reconstruction he had in fact paid it, into the wrong account, while the court’s records still showed him a debtor. So his name went onto the Supreme Court’s roster of “dishonest persons,” and a standing rule did the rest: every agency that touches that roster refuses the listed the same things. No villain at a screen. A clerk’s list, wired between offices, and a policy that turned the list into a wall.
The machine that erased Liu Hu from the planes and trains was not science fiction. It was paperwork, networked, with no door back out: a list, a wire, and a rule. In 2018 the agency that runs it reported that people on the list were blocked from buying plane tickets roughly seventeen million times in a single year, and high-speed-rail tickets five million more.
You think this is a story about China. It is a story about a method, the method a country reaches for when it has decided that some of its people are no longer worth needing. And it is being assembled, part by part, in the one that says it is racing to defeat it.
The cheerful version
You can hear the pitch most clearly from the most cheerful man in the room.
In June, the futurist Peter Diamandis — X Prize founder, Singularity University co-founder, the closest thing the abundance gospel has to a high priest — published an essay welcoming what he calls a “trillion-sensor electric skin” spreading over the world. He does not call this surveillance. He calls it sunlight. “Sunlight,” he writes, “is becoming infrastructure.”
Give him his due, because the honest core of his case is real. When the cameras go on, people do behave better. He has the receipts: a controlled trial in Rialto, California, where police body cameras cut use-of-force by sixty percent and citizen complaints by nearly ninety; four decades of studies finding that CCTV nudges crime down; the commercial satellite images that caught Russia massing on the Ukrainian border before the denials could be issued. Being watched civilizes. That part is true, and it is the strongest thing he says.
And then Diamandis does something most salesmen are trained never to do. He names the trap himself. A surveillance state and an accountable democracy, he writes, “can run on the identical sensor network.” The world he worries about “isn’t the one where everyone is visible. It’s the one where the powerful can see everyone, while no one can see them. Transparency only builds trust when it points both ways.”
He writes that sentence, and then he steps over it. He gestures at “fighting for a world where the watching goes both ways,” signs off “To a future of Abundance,” and invites the reader — whom he addresses, by name, as an investor — to a longevity retreat where the real draw is the pre-market deal flow. He sees the cage clearly enough to describe the lock, and walks in anyway, because he has done the arithmetic and concluded he will be the one holding the key. The people who own the cameras welcome a world of total visibility because they expect the watching to land on everyone else. Diamandis is the specimen here because he is the most honest, he says the quiet part and the loud part in the same paragraph and never notices the contradiction, because from where he sits there isn’t one. When the optimism falls away you get Larry Ellison, who told a room of his own investors the same thing without the abundance wrapping: citizens, he promised them, “will be on their best behavior because we are constantly recording and reporting everything that’s going on.” One man sells it as sunlight. The other sells it as a product. It is the same machine.
The slogan, said plainly
Notice the frame the whole thing arrives in. We are told, constantly, that this build-out is how we, the United States, win. That the trillion in data centers and the scramble for chips and the warnings that we cannot afford to lose all come down to one thing: the United States must beat China, so that the free and open world prevails over the surveillance state. We race so the cameras don’t win.
And then the same men build the cameras. Palantir wired into the IRS and the deportation agency. Facial recognition pushed down to local police. An immigration service holding the driver’s-license photos of most adults in the country. The instrument we are told we need in order to defeat the surveillance state turns out, on inspection, to be a surveillance state.
We want to beat China so we can become China.
The United States is not the People’s Republic. We still have courts, a vote, a press, a First Amendment and those are the only thing that ever stood between us and Liu Hu’s screen, and they are being cut, in the same years, by the very people building the cameras. The press that exposes the machine? This administration named an AI company a national-security threat for talking to that press, and a federal judge called the move “Orwellian.” The intelligence oversight meant to limit warrantless surveillance? Threatened from the inside, by who gets installed to run it. The brakes are being cut by the people telling you the gas pedal is patriotism.
Which way the glass faces
A surveillance state is never shared out evenly, and that is the part the pitch is built to hide. Watching has a direction.
Think of a driver under race control. Every camera angle is on him; telemetry logs every input his hands and feet make; a booth of stewards he will never meet can study all of it and hand down a penalty in seconds. He drives clean because he is watched, exactly as Diamandis promises. But the watching runs one way. The driver cannot see in. He cannot audit how it decided, cannot cross-examine it, often cannot even get a straight account of the rule he supposedly broke. He is the most visible man on the circuit and the booth is made of two way mirrors. That asymmetry — total visibility one way, mirrored glass the other, is not a flaw in the system. It is the system. Watch you always; allow those on the outside to see themselves and others, but those on the inside, watch us never.
Look at how much of you is already outside the booth. Diamandis lays the apparatus out top to bottom. It starts in orbit, where Planet Labs satellites photograph every square meter of the Earth’s surface every day and tens of thousands of Starlink relays blanket the sky. Beneath them, a million drones and air taxis coming in the next few years, each a moving camera. On the street, a single Waymo carries thirteen cameras, four lidar, and six radar and throws off four terabytes of data an hour. One car, four terabytes, every hour it rolls. The humanoid robots Musk and others are building, ten billion of them by his count, arrive “with eyes.” A billion fixed cameras already watch the doorways and the corners. Seven billion phones ride in seven billion pockets, each a microphone and a camera and a beacon that never fully sleeps. Forty billion connected devices by 2030, he says, a skin of sensors drawn over the planet from the stratosphere to the front step.
And he stops one layer short. The last sensor is not on the street or the doorframe. It is on your wrist, reading your heart rate as you read this, along with your sleep, the oxygen in your blood, the gap between one beat and the next; some of them now run an ECG. The stack that opens with a satellite photographing a continent closes on a band of plastic that knows your pulse before you do.
Now ask what all of it is all for. A sensor has one job: to report. The satellite reports the field, the Waymo reports the street, the phone reports the room, the watch reports the body. Every one of those reports travels the same direction, up and inward, to whoever owns the layer. Pooled, they stop being readings and become a portrait: where you were, who you were beside, what you said near the speaker, how you slept, the moment your heart jumped and what was on the screen when it did. The electric skin is one instrument with a trillion nerve endings, and you are not the hand it answers to.
So the only question that has ever mattered about surveillance is the one nobody selling it wants you to ask: which way does the two way mirror face? A network that lets the public watch the powerful is accountability, actually, it’s democracy, it is the body camera that convicts the lying cop, the satellite that catches the massing army. A network that lets the powerful watch the public is the thing that erased Liu Hu from the trains. It is the same hardware. The only variable is the direction, and the direction is being chosen for you.
Diamandis calls it sunlight, and the old saying is that sunlight is the best disinfectant. So it is, when it falls on everyone. A light the many cannot step out of, with the few standing behind it, disinfects nothing. It’s an interrogation lamp.
The wall, this week
You do not have to imagine the American version. You can watch the wall going up in realtime now.
This month the gate stopped being a metaphor. In early June an executive order set up a process where the government gets to review the most powerful new models. “Covered frontier models,” in the order’s language, for up to a month before anyone else can have them, and helps decide which “trusted partners” get them first. The order is careful to say it creates no mandatory license; the lawyers will tell you, correctly, that it bans one. What it builds instead is a softer thing that arrives at the same place: a review the labs will not skip, and a list of the blessed they are allowed to work with. Days later a separate export-control order reached into Anthropic and switched off two of its best models, Fable 5 and Mythos 5, for every foreign national on earth, including the company’s own foreign-born engineers. Then, late in the month, the Commerce Department partly reopened the gate, Mythos 5 could go to a hand-picked roster of about a hundred American institutions, listed by name in an annex while Fable stayed dark. OpenAI’s newest model, the one it calls Sol, launched the same week to roughly twenty government-vetted partners, a hidden list, and no one else. These are the names and lists of one week in late June 2026, and they will have moved by the time you read this. The shape will not.
Liu Hu’s story and the machinery described is identical; only the product has changed. A list of the approved. A wire between the government and the firms. A rule about who is refused.
And notice who that leaves on the inside. The frontier goes to a hundred institutions with the right relationships; everyone else trails behind by however many months the review takes, on whatever the labs choose to ship to the public. The people who decide who makes the list gain, with each cycle, more of the only thing that matters here, the power to say who gets the future and who waits for it. Wes Roth, a YouTuber who covers the industry for an audience of AI enthusiasts, no critic of the technology, no stake in my politics, watched the same news and landed on the phrase I would: a permanent underclass. The penthouse gets superintelligence. The floors below get the version from a few months ago and a place in line. The arrangement raises a question older than any of this hardware: what a society does with the people it has decided it no longer needs?
How do you decide who is allowed through the gate? You have to know who everyone is. The most-discussed answer in the industry right now, floated, this week, by the same optimists selling the abundance, is to require proof that you are a specific, unique human before the powerful models will open for you: an identity check, and the cleanest version on offer is the one Sam Altman has spent years building, the eye-scanning orb that enrolls your iris and hands back a token certifying that you are you. Follow that to its end. To be permitted the tool, you submit to the scan. The surveillance stops being the price you pay for safety and becomes the turnstile you pass through to be allowed upstairs. You are watched in order to be admitted, and the people who own the turnstile are never on the far side of it.
And the enforcement, if it comes, should end the argument about whether “become China” is hyperbole. The proposals already circulating for keeping powerful open models out of the wrong hands read like a translation of the Chinese playbook: block the websites that host the model weights, lean on the code repositories to pull them down, log the addresses that downloaded them, make an example of a few people with a raid loud enough to frighten the rest. That is not a slippery-slope worry. That is a description, written by people who want it, of a Great Firewall for intelligence. Proposed, in earnest, by the country that justifies the entire AI project as the war against exactly that.
The gate, at least, has produced the first thing everyone agrees on in years: safety people and accelerationists, libertarians and regulators all hated the two-tier frontier in the same breath, and a wall this visible is one you can still stop. The real danger is the part that happens where you can’t see it.
And it is learning to build itself
Because the thing behind the wall is no longer holding still.
For two years the gap between the best model inside a lab and the best one you could use was only a few months wide, and that narrowness was a kind of safety: it meant no one could get very far ahead in secret. That is the assumption now dissolving. The most-watched number in the field this spring came from Anthropic’s own co-founder, Jack Clark, who put better-than-even odds, sixty percent and up, on AI research with no human in the loop arriving before the end of 2028. A model, in other words, capable enough to train its successor, which trains its successor, the loop tightening past the speed of the people meant to supervise it. Demis Hassabis, who runs Google DeepMind, says every serious lab is now racing toward it.
This is not a forecast floating free of evidence. Inside Anthropic, more than eighty percent of the code shipped is already written by the model; a typical engineer now merges eight times what they did two years ago. On a new benchmark called MirrorCode, which hides a program’s source and asks the AI to rebuild it from its behavior alone, a leading model reconstructed a sixteen-thousand-line scientific toolkit that would take a human engineer 6-8 weeks on average, and did it in fourteen hours, for $250. One task on that benchmark ran for nineteen days without a human touching it.
Which is what turns a temporary head start into a permanent distance. A capability that improves itself, accelerating, behind a gate that admits a hundred names and turns the rest away, stops being a head start and becomes a caste, the tool on the inside building the next tool, the gap that was a few months wide opening without limit while the public sees only what the labs choose to ship. The labs, meanwhile, use the self-improving models freely on themselves and write terms forbidding everyone else from using them to build a competitor, to protect, they say, the national lead. The one real attempt to pry the loop open is already funded by Andreessen Horowitz and Nvidia, the very concentration the escape was supposed to route around.
It’s the two way mirror glass again, turned the other way, no longer only about who may watch whom, but about who is permitted to know what is being built. We could once see, roughly, how fast the frontier was moving, because it moved in public. Behind a review wall, with a model improving itself on the inside, that visibility is gone. It is now genuinely possible for a lab to cross a line that matters. To build something far past anything you have been allowed to touch and for you to learn of it afterward, if at all. The most consequential thing happening in the world would be happening on the far side of the two way mirror.
The Unwatched
There is a particular stratum doing the choosing. Not the merely rich, a specific slice of them, the people who own the substrate of the information economy, the compute and the models and the data and the watching tools themselves, and who have turned that ownership into a claim: that building the machines which run the world entitles them to a different relationship with the rules than the rest of us get. Call them The Unwatched. You do not have to infer what they believe. They have published it.
In 2009 Peter Thiel wrote, in an essay he signed, “I no longer believe that freedom and democracy are compatible.” The sentence everyone quotes. The one that matters more is the next move: the task, he wrote, is “how to escape not via politics but beyond it.” Not reform democracy, escape it, into the spaces a vote can’t reach. Balaji Srinivasan turned that instinct into a program he calls the network state, a sovereign community you build and own and seek recognition for, pitched in a talk he titled, without irony, “Silicon Valley’s Ultimate Exit.” And Curtis Yarvin supplied the theory underneath, the one that wants to retire the entire civil service and install a national chief executive, and that asks Americans, in his words, to “get over their dictator phobia.”
You could file all of that under fringe, until you notice where it landed. The Vice President of the United States has cited Curtis Yarvin by name. Before the office, he described the advice he would give a returning president: fire every single mid-level bureaucrat, replace them with our people, and when the courts try to stop you, say “the chief justice has made his ruling; now let him enforce it.” The blog post became the sitting Vice President’s theory of power. This is no longer something strange men write on the internet. It governs.
Who pays
Go back to the screen.
In January 2020 a man named Robert Williams was arrested on the front lawn of his house outside Detroit, in front of his wife and his two small daughters, and held for roughly thirty hours. He had done nothing. A facial-recognition algorithm had matched his driver’s-license photo to grainy footage of a shoplifter, and the system trusted the match more than it trusted him. There was no list he could appeal, because he never knew he was on one until the handcuffs were on. He is among the first Americans known to be wrongfully arrested by a face match. He will not be the last, because the system Diamandis calls sunlight makes exactly that match, at exactly that scale, the ordinary case.
Robert Williams could not opt out. There was no door he could buy. And the man who wrote that the answer is to escape democracy holds a New Zealand passport. That is watch-you-always and watch-us-never, made flesh: the camera owns the man it misidentified, and the man who theorized the exit has already bought one.
The American version of Liu Hu’s wall does not need to be invented, and it does not need a science-fiction score. We are building it in plain sight. We already keep no-fly lists and watchlists no one can see or contest. We already let one immigration agency hold most adults’ license photos. We are already paying Palantir to wire the databases into each other. The only thing China added to a stack of parts we already own was the rule that everyone who touches the list refuses you the same things. We have the list. We are laying the wire. The rule is one policy away.
We don’t need you anymore
The man who runs the government has told you, more than once, which era he wants back.
“We were at our richest from 1870 to 1913,” the President said within days of taking office. “That’s when we were a tariff country.” In his inaugural he praised William McKinley, the president of that age, as a “great president” who “made our country very rich.” He has been candid about the destination. That stretch, 1870 to roughly 1900, already has a name, and it was not meant as a compliment. It is the Gilded Age: a thin skin of gold over a great deal of rot, the years of Rockefeller and Morgan and Carnegie and Vanderbilt, when the top one percent held something like forty percent of the nation’s wealth, when the men who owned the railroads owned the senators too, and when, in the Panic of 1893, a quarter of the country was thrown out of work. It looked magnificent if you stood where the barons stood. From anywhere else it was a furnace. I spent four essays once on why no one standing outside the penthouse should want it back.
The first Gilded Age ended. And it ended for a reason that has nothing to do with the conscience of anyone at the top. It ended because the barons still needed the people. The steel did not pour itself; the rails did not lay themselves; the mills could not run without the bodies on the floor. That need was the only leverage the powerless have ever had, and they used it. They organized, they struck, they bled at Homestead and Pullman, until the country wrote the need into law. The top tax rate climbed to ninety-one percent. The share held at the very top was cut roughly in half. The thing we now call the middle class was built, deliberately, as the price of needing the workers who built everything else. They shared because they could not run the machine without you.
Now read what this government is doing.
It is rebuilding the Gilded Age concentration, openly, as a stated goal. But a Gilded Age economy is still a consumption economy: it runs on people with enough in their pockets to buy what gets made. You cannot keep your customers and gut them in the same breath. And yet that is exactly what is happening. The same months that blessed a half-trillion-dollar monument to the machines, Stargate, “the largest AI infrastructure project in history,” announced from the White House on the second day of the new term, took the better part of a trillion dollars out of Medicaid, cut food assistance more deeply than at any point in its history, canceled research grants by the thousand and half a billion dollars of work on the vaccines for the next pandemic, and let a subsidy lapse so an ordinary family’s insurance premium more than doubled. Courts have clawed some of it back; the government answered by deleting the records of what it cut. But gutting your own consumer base is not Gilded Age economics. It is not any economics at all, unless you have concluded that you no longer need consumers. That is the confession folded into the budget. You do not cut the people this way while pouring everything into the machines unless you have already run the arithmetic and decided that, in the economy you are building, the people are neither the labor nor the market. They are overhead.
And this is why this time is not the last time. That earlier collapse came because the leverage survived, because the need came back, and the people still held it. This one is being built on the single technology designed to make sure it never does. Every machine before it displaced one kind of worker and then needed a new kind to run it; there was always a next field to walk to. A general intelligence is the first that doesn’t. It is not a better loom that still wants a weaver. It is the weaver. The project is not only to concentrate the wealth. It is to retire the leverage before it can be used, to climb back to the Gilded Age and then pull up the ladder the first one was forced to lower.
Once you have decided you no longer need the people, you inherit the problem the robber barons never solved: a population you have discarded is dangerous. Diamandis, the optimist, names it himself, “the most dangerous demographic in history,” the young and capable and shut out, the exact cohort that has brought down regimes before. The first Gilded Age ended in reform; it could have ended in revolution, and the men at the top knew it. So what do you build, when you have made a population surplus and cannot afford to let it organize the way it did at Homestead? Total visibility, pointed down. The gate that decides who is still worth including. The watchlists, the face scans, the civilizing sunlight. The surveillance state is not a side effect of the new economy. It is the new economy’s answer to the question the old one answered by sharing: how do you hold a population you have stopped needing. Call it, at last, The Watchers’ bargain. The rich accept the cameras because they have found a way to stop needing you, and the cameras are how you keep the people you no longer need.
So go back, one last time, to Liu Hu at the ticket screen. Not qualified. No file, no warrant, no notice, no one to call. He was not refused by a machine that needed him. He was refused by a system that had filed him under no-longer-required, and there is no appeal from that, because it is not an error. It is a verdict. We have a word for everyone left on the other side of the two way mirror: schmucks.
Read through the budget and the buildout and it is colder than that. A status, an entry in a ledger, the gold leaf of a new Gilded Age laid over the people it has quietly decided it can do without. They are telling you, in everything they fund and everything they cut, that they already know they don’t need you. Which direction does the sunlight point? Down. It was always going to point down. The one thing the first Gilded Age settled, and this one is built to never have to: the discarded are powerless right up until the moment they remember they were the leverage all along. Everything being built right now, the gate, the cameras, the ledger that sorts the needed from the overhead is being built to make sure that moment comes too late.
Time to break the interrogation lamp since it’s not the sun.
Sources
China’s blacklist and Liu Hu: The Globe and Mail; the “there is no single national score” correction: China Law Translate; the seventeen-million-blocked-purchases figure: MIT Technology Review.
Peter Diamandis, “Visibility, Transparency & Trust” — the trillion-sensor “electric skin,” the body-cam and CCTV data, and the caveat he names and steps over: Abundance / Metatrends.
Larry Ellison, “citizens will be on their best behavior” (Oracle Financial Analysts Meeting, September 2024): TechCrunch.
The June 2026 frontier-model gate — the executive order: The White House; the Fable 5 / Mythos 5 export shutdown: Forbes; Mythos 5 reopened to about a hundred institutions: CNBC; GPT-5.6 “Sol” limited to trusted partners: CNBC.
Recursive self-improvement — Jack Clark’s sixty-percent-by-2028: Import AI #455; the MirrorCode benchmark: Epoch AI; the $200M raise behind the one attempt to open the loop: Unite.AI.
Peter Thiel, “I no longer believe that freedom and democracy are compatible”: Cato Unbound; Balaji Srinivasan, The Network State: thenetworkstate.com; JD Vance’s “fire every single mid-level bureaucrat… now let him enforce it,” and his citation of Curtis Yarvin: Defense One.
Robert Williams, wrongfully arrested by a facial-recognition match: The New York Times.
Stargate, the half-trillion-dollar announcement: SoftBank; how little of it is actually built: Epoch AI.
Trump on 1870–1913 as America’s “richest” era, and his praise of McKinley: PBS NewsHour.
The 2025 reconciliation law — the Medicaid and SNAP cuts and the coverage loss: KFF; the cancelled mRNA-vaccine contracts: STAT.
My four-part case against a new Gilded Age: redkrel.substack.com.