U.S. government shuts down Anthropic's two most powerful AI models over national security concerns, a move the company argues will chill AI development industry-wide.

What happened

The U.S. government ordered Anthropic on Friday to immediately disable Claude Mythos 5 and Claude Fable 5 worldwide, citing national security concerns framed as an export control action. Anthropic complied but stated in a blog post that the government's underlying concern is a claimed jailbreak of Fable 5 — specifically, a method of prompting the model to read code and identify software flaws. Mythos 5, previewed in early April, was kept tightly restricted through a controlled program called Project Glasswing shared with roughly 50 vetted organizations including Amazon, Apple, Google, Microsoft, and CrowdStrike. Fable 5 was released just three days before the shutdown and was immediately the most capable AI model available to the public, according to benchmark tests from Vals AI.

Why it matters

Anthropic has built its public identity on being the safety-conscious alternative to rivals, and has staked much of its messaging on warning the world that Mythos was uniquely dangerous — so dangerous it could not be released publicly. The irony is that this very caution in restricting Mythos appears to have drawn the government scrutiny that now disrupts the company's business. The company's broader frustration is clear: it argues that the claimed jailbreak represents a capability already widely available in publicly accessible models including OpenAI's GPT-5.5, and that applying the government's standard across the industry would halt all new model deployments. Anthropic is widely expected to pursue an IPO this year.

What to watch

Anthropic asserts that its strongest safeguards operate through independent classifier systems separate from the model itself, meaning that even if someone convinces Fable to continue responding past a refusal, underlying protections against the most dangerous outputs remain in place. The company disputes that a narrow, non-universal jailbreak should trigger a recall of a commercial model deployed to hundreds of millions of people.