Friday morning, Anthropic woke up to a government order: kill access to Fable 5 and Mythos 5. Now. All customers locked out. Foreign nationals banned entirely, even if they work there. The company complied within hours.
This isn't a quiet policy shift. This is the US government essentially saying "we don't trust you to keep this safe" — and doing it fast enough that Anthropic didn't even get time to argue before pulling the plug. The two models are gone from production. Customers trying to access them get a wall.
Here's what actually happened: Anthropic launched Fable on June 9 as a public version of Mythos, its classified cybersecurity model. Fable was built to do things no previous Anthropic model could do. In testing, it beat Pokémon FireRed. Claude couldn't beat the original. That's the gap we're talking about. The company loaded it with safeguards specifically designed to prevent misuse — so many safeguards that users complained they were too strict. Then someone found a way around them.
The government won't say who. A verbal tip came in about a jailbreak method. One method. Not confirmed. Not deployed. Just theoretical enough that the feds decided this warranted a total shutdown of both models for every single customer on Earth. Anthropic gets to explain more in 24 hours, but by then the damage is done — the models are already offline.
Anthropic's response is pointed: they disagree with this. The company has actually been pushing for more AI regulation, but they wanted it to be "transparent, fair, clear, and grounded in technical facts." This move hits none of those marks. A potential vulnerability became a reason to nuke a product nobody's actually proven is broken. The company called it out directly: "This action does not adhere to those principles."
What happens next depends on what Anthropic reveals in the next 24 hours. If the jailbreak is real and dangerous, the government looks justified. If it's theoretical or overblown, Anthropic has a PR nightmare and a serious trust problem with Washington. Either way, every AI company is watching to see if a single tip can now kill your product overnight.