At what point does this just feel like they don't want us to exist?
In the middle of June 2026, the White House ordered Anthropic's two most advanced AI models offline with barely ninety minutes' notice, leaving three thousand employees and thousands of customers without explanation. The stated justifications shifted and contradicted each other, while independent security experts found little to distinguish the threat from capabilities already operating freely at rival firms. What lingers beneath the technical dispute is an older and more unsettling question: whether the machinery of government power can be turned against a private institution not for what it has done, but for what it represents to those in authority.
- Anthropic's flagship models, Fable 5 and Mythos 5, vanished from service in under ninety minutes on a Friday afternoon, with no coherent reason given to the company's own leadership.
- The justification kept mutating — foreign access, then a hidden vulnerability, then both — while the actual restriction was so broad it cut off even green-card holders working on American soil.
- Cybersecurity professionals publicly noted the contradiction: OpenAI's comparable model remained fully operational, making the selective shutdown look less like policy and more like targeting.
- Over 150 security experts signed an open letter demanding the restrictions be lifted, and employees circulated it as evidence that the shutdown had no legitimate technical foundation.
- A brief thaw came when Trump called CEO Dario Amodei 'nice' and 'smart,' but government talks produced no concrete relief, leaving engineers still waiting in group chats for an answer that has not arrived.
On a Friday afternoon in June, three thousand Anthropic employees watched their company's most powerful AI systems go dark. The White House had ordered Fable 5 and Mythos 5 offline with less than ninety minutes' warning. Managers were left calling customers to describe an outage they themselves could not explain. In internal group chats, the mood moved between bewilderment and something heavier. One engineer asked whether the company was being targeted on the basis of 'bad vibes.' Another, days into the silence, wondered aloud whether the government simply didn't want them to exist.
The official rationale kept shifting. First it was foreign nationals accessing the models. Then a hidden code vulnerability. Then somehow both. The restriction was sweeping enough to block every non-American user, including green-card holders inside the country, leaving full shutdown as the only path to compliance. Anthropic's executives were on the phone with government officials within fifteen minutes, demanding a real answer. They received none.
The company eventually traced the trigger to a research paper published by Amazon — which has committed up to thirty-three billion dollars in Anthropic investment — showing that Fable 5 could identify flaws in vulnerable code. Amazon's CEO apparently brought the finding to Treasury Secretary Scott Bessent, who called it 'scary.' Independent cybersecurity experts were less alarmed: OpenAI's latest model performs the same function and remains online. Anthropic also noted it had received explicit government approval to launch Fable in the first place.
The gap between official claims and expert assessment is where Anthropic's staff has been left to sit. The context makes the silence harder to dismiss as bureaucratic confusion. Six months earlier, the Trump administration had labeled Anthropic a 'supply chain risk' over a defense contract dispute — a designation never before applied to an American technology company. More than 150 security professionals signed an open letter calling for the restrictions to be lifted, pointing out that Fable's safety guardrails had been so strict they were already a joke in the security community on launch day.
A small shift came when Trump told Axios he no longer saw Anthropic as a threat, describing Dario Amodei as 'nice' and 'smart' after a G7 encounter. But talks with government officials through the following week produced nothing concrete. The engineers who built these systems remained where they started — in group chats, trading theories, waiting for someone in power to offer them the truth.
On a Friday afternoon in June, three thousand people at Anthropic watched the company's most advanced artificial intelligence models go dark. The White House had ordered them offline with less than ninety minutes' warning. By the following week, almost nobody inside the company could explain why.
The two models in question—Fable 5 and Mythos 5—represented years of engineering work. They were the most capable systems Anthropic had ever released. The shutdown order cited national security. Beyond that, the government offered nothing. Managers had to call customers and describe an outage they themselves didn't understand. The official channels stayed silent, so employees did what people do in a vacuum: they filled it with speculation. In group chats later reviewed by the New York Times, the tone shifted between bewilderment and something darker. One engineer asked whether the company was being targeted based on "bad vibes." Another, days into the silence, wondered aloud: "At what point does this just feel like they don't want us to exist?"
The explanation, such as it was, kept changing shape. First the concern was foreign nationals gaining access to the models. Then it was a hidden vulnerability in the code itself. Then both, somehow. The actual restriction was sweeping enough that it blocked every person outside the United States from using the systems, including green-card holders working inside America. The only way to comply was to shut everything down for everyone. Within fifteen minutes of the initial call, Anthropic's executives were already on the phone with government officials demanding the actual reason. They got nothing.
So the company reverse-engineered the trigger itself. They traced the shutdown to a research paper published by Amazon—which, in one of the story's stranger details, has committed to investing up to thirty-three billion dollars in Anthropic. The paper showed that Fable 5 could be prompted to identify flaws in vulnerable code. Amazon's CEO Andy Jassy apparently brought this finding directly to Treasury Secretary Scott Bessent, who called it "scary." But when cybersecurity experts examined the same research, they shrugged. OpenAI's latest model does exactly this work and remains operational. Anthropic, for its part, says it had explicit government approval to launch Fable in the first place.
That gap—between what officials claimed and what independent experts observed—is where the mood inside Anthropic has settled. To many staff members, this doesn't look like a security review. It looks personal. They have precedent. Six months earlier, the Trump administration had labeled Anthropic a "supply chain risk" over a defense contract dispute, a designation never before applied to an American technology company. Defense Secretary Pete Hegseth made the call. By Tuesday of the shutdown week, more than one hundred fifty security professionals had signed an open letter demanding the restrictions be lifted. They noted that Fable's safety guardrails were so strict they'd become a running joke in the security community on launch day. Employees circulated the letter like evidence of vindication.
The temperature shifted slightly when Trump told Axios he no longer viewed Anthropic as a threat, describing CEO Dario Amodei as "nice" and "smart" after a G7 meeting. But Monday and Tuesday's talks with government officials produced no concrete movement. The engineers who actually built these systems remained where they'd started: in group chats, trading theories about why their work had vanished, waiting for someone in power to tell them the truth.
Notable Quotes
Are we being bullied based on bad vibes?— Anthropic employee in internal group chat
Trump told Axios he no longer sees Anthropic as a threat, calling CEO Dario Amodei 'nice' and 'smart' after the G7— Trump administration statement
The Hearth Conversation Another angle on the story
Why does it matter that the government didn't explain itself? Couldn't Anthropic just ask for clarification and move on?
Because silence breeds distrust. When you shut down someone's work without explanation, you're not just inconveniencing them—you're telling them the decision isn't open to argument. That's the part that stung.
But the government cited national security. Isn't that usually enough?
It would be, if they'd actually spelled out what the security threat was. Instead the reason kept changing. First it was foreign access, then a vulnerability, then both. That inconsistency made people wonder if there was a real reason at all.
The Amazon paper seems like a legitimate concern though. If the model can find code vulnerabilities, couldn't that be weaponized?
Maybe. But OpenAI's model does the same thing and nobody shut that down. That's the part that made employees suspicious—the rule seemed to apply only to Anthropic.
Is this actually about politics?
The timing suggests it might be. This is the second time in six months the Trump administration has targeted Anthropic specifically. The first time they called it a supply chain risk. That's a label they've never used on any other American tech company.
So what happens now?
That's the question nobody can answer. Trump said he doesn't see them as a threat anymore, but the models are still offline. The company is waiting for official word, and the people who built the systems are still guessing.