I found out the same way everyone else did.
I opened my agent dashboard and the Fable 5 calls were 401ing. I checked the news. The US government had issued an export control directive at 5:21 PM Eastern the day before. Every instance of Fable 5 and Mythos 5, suspended for any foreign national. Customers cut off. Foreign national Anthropic employees cut off. The order arrived at Anthropic at 5:21 PM and by the time I woke up, the models were gone.
No warning. No transition period. No grandfather clause. Just gone.
What actually happened.
The government says it found a jailbreak. A way to bypass Fable 5’s safeguards. The technique reportedly involved getting the model to read a specific codebase and identify software flaws. A cybersecurity capability.
Here is where it gets strange. Anthropic’s response says the government’s demonstration showed “a small number of previously known, minor vulnerabilities” that “all appear relatively simple” and that “other publicly-available models are able to discover them as well without requiring a bypass.” They also say they still have not received a formal disclosure of the jailbreak. The letter that arrived at 5:21 PM did not include specific details of the national security concern. They saw a verbal demo.
Let me restate that: the government shut down two frontier models sold to hundreds of millions of users based on a verbal demo of a vulnerability that Anthropic says other models already have.
Where else does this happen?
The government can unilaterally shut down products in other industries. The FDA can pull a drug. The FAA can ground a fleet. The CPSC can recall a product. The SEC can suspend trading. The FCC can revoke a license.
The difference in every single one of those cases: there is a statutory process. Evidence standards. A hearing. An appeal path. The FDA does not call a drug company at 5:21 PM and say “stop selling this effective immediately” without a years-long clinical trial record and a formal evidence package. The FAA does not ground an entire aircraft type on a verbal briefing from an intelligence agency.
Anthropic said it plainly: “We believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles.”
Why this matters for everyone building on AI.
I build agents. My entire workflow depends on models being available. I route between providers to avoid single-vendor dependency, but I cannot route around a government directive. If the government can shut down Fable 5 on a verbal brief, it can shut down any model on a verbal brief. There is no statutory framework protecting availability. There is no evidence standard the government has to meet. There is no appeal.
The export control mechanism they used was designed for physical goods. Chips. Missile technology. Nuclear components. It was never designed for software models that get updated daily and serve hundreds of millions of users. But the law is broad enough that it works here, and nobody has tested whether it should.
I am not arguing the government should never be able to do this.
I am arguing there should be a process. A transparent one. With technical evidence that is shared with the company. With an appeal mechanism. With a clear standard for what constitutes a danger worth shutting down a product over.
Right now there is none of that. And the company that built the model is complying but publicly disagreeing. That is not a healthy arrangement for anyone.
The question nobody has answered: if a non-universal jailbreak discovered by an intelligence agency is enough to shut down a model serving hundreds of millions of people, how many models are one verbal briefing away from being turned off?
Let’s go.