The 5:21 Email
Nothing good has ever arrived at 5:21 on a Friday afternoon. Ask anyone who has run a support queue. The late-Friday message is always the one that eats your weekend.
On June 12, at exactly 5:21pm Eastern, Anthropic got one of those. It came from the US Commerce Department, signed by Secretary Howard Lutnick, and it carried an export control directive: no foreign national, whether inside the United States or out, could use the company’s two newest models, Fable 5 and Mythos 5. That included Anthropic’s own non-American employees. Happy Friday.
When You Can’t Check IDs, You Close the Bar
Here is where it gets interesting, and a little absurd. The order was about who could use the models, not whether they existed. In theory, Anthropic just had to keep foreign nationals out.
In practice, it had no reliable way to check everyone’s passport at the door. A chatbot does not card you. So faced with letting the right people in and nobody else, the company did the only clean thing available: it shut the whole place down. Both models, every customer, every region, even on AWS.
A rule meant for some people became a blackout for all people, because the system had no good way to ask the question at the entrance. If you have ever locked an entire feature because you could not safely scope it to the few users who needed it, you already understand the engineering here. When you cannot tell who is who, denying everyone is the boring, defensible answer.
Careful What You Build the Lever For
Now the part that made me put my Coke Zero down. Two days before the letter, on June 10, Anthropic’s own CEO published a piece arguing that the government should be able to step in and block unsafe AI deployments. He even named Mythos as the prime example of the risk.
Forty-eight hours later, the government picked up that exact lever and pulled it. Anthropic’s response, roughly, was that yes, the government should have this power, but not like this. The order was vague, and the cited jailbreak was a narrow one (it amounted to asking the model to read some code and fix the bugs, which is, you know, the entire job). Similar tricks already work on other models. Asking for a tool and then being startled by its grip is one of the oldest stories there is.
The Tip From Inside the House
There is a quieter thread worth pulling. Reporting from the Wall Street Journal suggests the Friday letter was prompted by concerns Amazon’s CEO raised with the White House, while Axios reports a rival company claimed it had cracked the model.
Sit with that for a second. Amazon is one of Anthropic’s biggest backers and the cloud provider hosting these very models. A nudge from that direction is not your competitor across town. It is a roommate quietly mentioning to the landlord that maybe your room should be locked. I am not saying that is what happened. I am saying the seating chart is interesting.
What Your Stack Is Actually Standing On
Strip away the politics and you are left with a very practical question for anyone building on top of someone else’s AI. One letter, sent at the worst possible minute on a Friday, made two state-of-the-art models disappear for everyone, everywhere, with no warning and no clear timeline back.
If your product, your workflow, or your weekend depends on a single model staying available, you are not standing on a platform. You are standing on a decision, made by people you will never meet, that can change at 5:21 on a Friday. Worth knowing where the floor is before it moves.
