Washington pulls Claude Fable 5 and Mythos from the market

June 12th, 2026, 5:21 PM ET. Anthropic shut down Claude Fable 5 & Mythos 5 globally at the request of the US government due to national security & export control reasons. Other models remain available, but their most potent models were taken offline. It’s disturbing to think that a tool people rely on for work got shut down by the government overnight without asking.

This is a brief overview of what happened. Mythos 5 is Anthropic’s strongest model. We saw a glimpse of it in April, but it was never released publicly. It was only accessible through a restricted program called Project Glasswing to ~50 hand-selected organizations (Amazon, Apple, Google, Microsoft, CrowdStrike, etc.) because it was so good at discovering software bugs. Fable 5 was the public release of the same model with added restrictions to prevent misuse in high-risk domains such as cybersecurity & biology. It was only live for 3 days before getting shutdown.

1The trigger is thinner than the consequence

They said that there was a “potential narrow, non-universal jailbreak” of Fable 5 which means that it was tricked into analyzing code and identifying bugs. Anthropic doesn’t mince words. They said they “disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.” They argue that other models such as OpenAI’s GPT-5.5 have similar abilities and that this type of code auditing is a common practice used by security experts daily.

It’s ironic that Sam Altman accused Anthropic of using “fear-based” marketing tactics to promote the dangers of Mythos earlier. Even if it’s unfair, it’s also true that Anthropic spent the last year hyping up the capabilities of this model, which probably caught the government’s attention. You can brag about how your AI is special for a whole year then complain when people with subpoena powers believe you.

If you’ve never used Mythos & never will, why does this matter? This could be the first time a cutting edge model has been recalled by a government while in use. If you depend on an AI assistant for your job, schoolwork, or personal projects, remember that you’re relying on a chain of suppliers you have no control over. Your model could become more restricted, change its behavior, or disappear completely from Tuesday to Friday, taking your workflow with it. These things are starting to feel like utilities, but they aren’t. They’re commercial goods operating within politics.

I suspect we will see more recalls like this in the future. As these models become stronger, they will start to resemble critical infrastructure that governments can take offline. It doesn’t mean we shouldn’t use them, but we shouldn’t pretend that they will last forever.