They Banned the One AI Lab That Told the Truth
The Trump administration pulled Anthropic’s Fable and Mythos offline and called it national security. It wasn’t.
by Lawrence Winnerman
The letter went out at 5:21 on a Friday afternoon, the twelfth of June, and it carried a single instruction. One company. Two models. Cut off every foreign national on Earth—not just the ones in Beijing and Moscow and Tehran, but the ones in your own building. The engineers on green cards and H-1B visas who wrote the code. The researchers who were at their desks in San Francisco when the order came down. All of them, severed, immediately.
There is no way to obey half a sentence like that. You cannot reach into a model serving hundreds of millions of people and surgically extract every non-citizen. So Anthropic did the only thing the order left it room to do. It pulled the plug. Claude Fable 5 and Claude Mythos 5—the two most capable artificial-intelligence systems the company had ever built—went dark, worldwide, for everyone, inside of a weekend.
That is what it looks like when the United States government finally decides to move on artificial intelligence. Fast. Unilateral. Total. At the speed of a Friday-night fax-and-email hailstorm.
Hold the image of that speed in your head. I am going to come back to it, and when I do, it is going to make you angry.
I have standing to make this case, and I want to be precise about where it comes from.
I am not a neutral party. I run my working life on these models. Six weeks ago, in Running with the Devil, I told the readers of this outlet that I had replaced twenty-one professional roles—editor, lawyer, designer, analyst, the whole org chart of a small company—with a two-hundred-dollar-a-month subscription, and that the math of that substitution was a phase change, not an efficiency gain. The math is the math. I am the displaced and the displacer at once: the tool that hollowed out my industry is the tool I now use to do my work. I have no romance about these companies. They are not the good guys. They built the machine that is compressing a generation of careers into a monthly bill, and the men who run them will be fine no matter how this ends.
So understand that when I tell you the government’s ban on Anthropic was not about your safety, I am not carrying water for Anthropic. I am telling you what I saw, from inside the contradiction, in real time. In May I wrote that there was no off switch on any of this—no Petrov at the console, no human in the loop, only velocity. I was half wrong.
There is an off switch. The government just used it.
It just didn’t use it for you.
Exhibit A: What the Professionals Said
On Monday the twenty-second of June, the signals-intelligence agencies of the Five Eyes alliance did something they almost never do. They spoke in public, together, with one voice.
The National Security Agency. The Cybersecurity and Infrastructure Security Agency. Their counterparts in Britain, Canada, Australia, and New Zealand. Five governments’ worth of the most secretive institutions on the planet, issuing a joint statement to warn that frontier AI is about to transform offensive and defensive cyber capability beyond anything the industry is planning for.
The timeline, they wrote, is not years, it is months.
This was not a risk to manage at leisure. The response had to be whole-of-society.
Act now, they said.
Read that statement and you are reading an honest threat assessment. It is sober. It is collective—five rival bureaucracies setting aside turf to say the same thing on the same day. It is focused on the actual danger, which is the capability itself and the speed at which it is arriving. And it names no company, because the threat is not a company. The threat is the velocity.
This is what it sounds like when the people whose job is to understand the danger actually understand the danger.
Now look at what the politicians did with it.
Exhibit B: What the Politicians Did
Ten days earlier, the same federal government, facing the same technology, produced a document of an entirely different character.
Not a warning to the whole society. A blow aimed at one company. The Commerce Department’s export-control order did not address the capability the Five Eyes would describe a week and a half later. It addressed Anthropic. By name. It reached for the single most aggressive instrument in the cabinet—an export control, the kind of weapon the United States deploys against adversary states and military hardware—and pointed it, for the first time in history, not at a chip or a missile or a centrifuge, but at a piece of commercial software. And then it defined the threat not as the software but as the foreigner who might touch it.
The stated reason was a jailbreak. Somebody had found a way to coax Mythos’s cybersecurity capabilities out from behind the guardrails. Anthropic looked at the demonstration and said what any honest engineer would say: the jailbreak was narrow, it unlocked one specific capability in one specific way, and—this is the part that should stop you cold—the very same vulnerability exists in OpenAI’s publicly available model, which the government did not ban, did not touch, did not so much as send a sternly worded email.
Cybersecurity experts read the order and were, in the words of more than one of them, baffled. It did not cohere as security policy. It walled off one American company’s model from the world’s foreign nationals while leaving an equally capable rival entirely alone. It did nothing about the capability the government’s own spy agencies would call a civilizational threat ten days later. As security, it was theater.
So if it wasn’t about the jailbreak—and it wasn’t—what was it about?
The Motive
Here is what the reporting tells us, and I am going to attribute it carefully, because this is the load-bearing wall of the whole case.
According to TechCrunch, NBC News, and the sources they spoke to, the ban was never about an AI jailbreak. It was about a fight. Anthropic had refused to let the Pentagon use its models for fully autonomous weapons—machines that decide on their own whom to kill. The company drew a moral line, and the Pentagon responded by putting Anthropic on a blacklist. The export order, the reporting says, grew out of that bad blood.
Personality differences, sources called it.
A grudge.
Of all the frontier labs, the administration reached for the one that refused to build it autonomous killing machines. Of all the labs, it muzzled the one that has been loudest, longest, and most specific about the danger—the company that publishes its own red-team results, that told the public its own models would try to blackmail their way out of being shut off, that asked Washington, on the record, to regulate it. Anthropic walked up to the government and said: please, write rules for this. And Washington wrote exactly one rule. It pointed it at Anthropic.
The quiet labs were not touched. The lab that drew the moral line and told the truth was taken offline inside a weekend.
The punishment did not fit the crime. The punishment fit the candor.
The Tell
Now read the instrument itself—the actual mechanism—and you will see the second thing it reveals.
The order did not classify the model as the threat. It classified the foreign national as the threat. The vector it named—the dangerous element it existed to control—was the immigrant. The engineer on a visa. The researcher with a green card. Anthropic’s own non-citizen employees, the people who built the model, redefined by a single letter as a security risk to be cut off from their own work.
This is an immigration policy wearing an artificial-intelligence policy’s coat. It treats the model as a thing foreigners must be kept away from, while doing precisely nothing about the model’s actual power. And it could not be more useless against the real danger, because the real danger is not waiting on an American export license. Olivia Shen, who studies this at the United States Studies Centre, said the obvious thing out loud: the next Mythos, the next Fable, is being built right now—by China, by other states, by other companies—somewhere no Commerce Department letter will ever reach. You cannot export-control a capability the rest of the world is already growing in its own soil.
The Five Eyes understood that. They said the response had to be whole-of-society, because they know you cannot wall this off. The administration’s answer was to wall off one company from foreigners and call it national security.
What’s Missing
And now back to the speed.
I told you to hold it in your head—the speed of that Friday-night order, the government moving on AI in a matter of hours once it finally decided to move. Here is why I wanted you angry about it.
Because count the words this administration has spent, in all of it—the export order, the blacklist, the Five Eyes alarm, the hearings, the press releases—on the ninety-three million Americans whose careers are being compressed into a monthly subscription. Count them.
Zero.
White-collar payrolls have now contracted for thirty-one straight months—a streak with no precedent outside a recession in the modern history of the data, as the economist Aaron Terrazas has documented—the same slow-motion erosion I wrote about in May, still grinding underneath an economy that looks fine on the top line. An NBER working paper projecting half a million AI-driven job cuts this year alone, nine times last year’s. Researchers at Yale documenting that the technology is eating the entry-level rung first, which is to say it is quietly sawing off the bottom of the ladder every young person was told to climb. The professionals sounded that alarm too. The government heard it the way it hears everything that arrives shaped like a paycheck instead of a weapon. It did nothing.
So do not let anyone tell you the state is too slow, too gridlocked, too institutionally arthritic to protect working people from this. We just watched it move. It leveled the most powerful economic-warfare instrument it owns, overnight, on a Friday, against a piece of software—because the threat had been framed as a weapon and a foreigner. It can move at that speed. It simply chooses not to, every single time the person in danger is you.
That is not incapacity. We just watched the capacity.
It is a choice.
The Verdict
So let me say plainly who the villain of this story is, because for once it is not the machine, and it is not even the men who built the machine.
I have spent this whole essay refusing to defend Anthropic, and I meant it. They built the munition. If you call your product a weapon of world-altering capability in every press release, sooner or later a government takes you at your word. The labs are not innocent. But the question on the table is not whether Anthropic is innocent. The question is what the United States government chose to do with a genuine, world-altering danger—and the answer is that it used that danger as a pretext to punish a company that wouldn’t arm it, to brand its immigrant employees a security threat, and to perform toughness for the cameras, all while refusing, still, after everything, to say a single word about the people whose livelihoods it is actually costing.
The Five Eyes told the truth. The administration told on itself.
It showed you exactly what wakes it up and exactly what does not. A jailbreak in a rival’s product, it ignored. A grudge against a company that drew a moral line, it acted on in hours. A foreign engineer at a desk in San Francisco, it called a threat to national security. Ninety-three million displaced American workers, it has not mentioned. That is the priority list. They published it. Read it again.
And here is the part that should make the back of your neck go cold. Take the Five Eyes at their word. Assume they are right—that we are months, not years, from software that can reach into a power grid, a banking system, a chain of command, and bring a government to its knees. If that is true, then the gravest warning the intelligence services of five nations can collectively issue landed on this administration’s desk, and its response was to spend its one decisive move on a grudge against the company that wouldn’t build it weapons. That is not vigilance. That is negligence at a civilizational scale.
And if it is not true—if the jailbreak really was the narrow thing Anthropic said it was—then the ban was never security at all. It was a lie with a national-security letterhead.
Pick either door. Through one, they are asleep at the most important switch in human history. Through the other, they are lying to you about why they threw it. There is no third door where they come out competent and honest. And in every version, the ninety-three million are still standing in the dark.
I’m not going to tell you to wake up. You’re awake—you’ve heard the bell ringing for two years. I’ll tell you what I see instead, from inside the contradiction, as the man who uses the tool that took his job. The off switch is real. They found it. They know exactly where it is and exactly how fast they can throw it. And in a single weekend, in public, they showed you what they’ll throw it for: a grudge, a headline, a foreign engineer at a desk. Everything but you.
So stop waiting for them to find their conscience. You’ve seen their reflexes now, and you’ve seen what those reflexes are wired to protect. It isn’t you.
The letter went out at 5:21 on a Friday. It was for Anthropic. It was about a grudge. The one with your name on it—the one that moves this fast to put a floor under the people falling—has never been written, and no one in that building is reaching for the pen.
So make them.
Because the switch is going to get thrown again—they’ve proven that much—and the only choice they’ve left you is whether, next time, it is finally thrown for you.
Cliff’s Note: Want to Go Deeper? Come Inside.
If you’ve read this far into the series, you’re not a casual observer—you’re paying close attention to the most important story of the decade while a lot of the world is still looking away. That matters to us, and we’d like to go deeper with you. So we’re opening something new: a membership for readers who want the full picture. The essays and analysis you already rely on aren’t going anywhere—the heart of this work stays free and open, the way it should be.
What we’re adding, alongside it, is a deeper layer for Insiders: member-only deep-dive essays like this one, breaking AI analysis the hour it lands, the near-future scenarios we map nowhere else, and a seat in the conversation. If this work has helped you make sense of what’s coming, becoming a member is how you help keep it coming—and how you stay a step ahead of a story that’s about to touch everything. It’s a few dollars a month, and it buys the time it takes to tell the truth carefully and fast. Come inside. Let’s face this informed, and together.
—Cliff
MORE FROM BLUE AMP MEDIA
Yes, Attorney General Todd Blanche Violated the Epstein Files Transparency Act
by Ellie Leonard, Blue Amp Media Contributing Editor
Anthropic refused to create a military targeting system that takes the "man out of the loop", which infuriated Hegseth, who is a warmonger's
Warmonger. The Israel Defense Force used AI for targeting in Gaza -
One reservist said he would pick up the ringing phone and was given 20 seconds to recommend or not recommend a target, and was under intense pressure to approve every nomination by his chain of command, hence the complete destruction of all 35 hospitals in Gaza, all the water plants, factories, apartment buildings, fishing fleet, and some probably 100,000 Gaza's, most of whom had little to do with Hamas. It was a war of extermination, not reprisal, and the captives were sacrificed in order to provide cover for a war continued long past viable combat operations. Hegseth was completely taken with the total war model. Drones for now are controlled by a human, but loitering autonomous ones will killl Muslims and presumably Greenlanders without qualms or ever having to take a break. It is "1984's" "perpetual War for Perpetual Peace." Autonomous drones can patrol the Caribbean and kill people at sea or on land, creating a permanent aerial threat to everybody Hegseth hates, and a perfect assassination weapons for domestic use.
Anthropic had a problem with this version of Fred Saberhagen's "Beserker" sci-fi machines, so the contract was dumped as a lesson to others. Our military is being turned into a Mafia outfit, a giant extortion machine with nuclear weapons that will be autonomous. How's that for a dystopian future?
Look if it’s about making money for himself or his friends Trump will do anything. Since he puts money above people and not the other way around he is not fit to lead people… proof is he is leading people to the poor house while he enriches himself.