U.S. orders Anthropic to shut off Claude Fable 5

U.S. orders – The U.S. government ordered Anthropic to immediately disable two of its most powerful AI models—Claude Fable 5 and Claude Mythos 5—citing national security concerns. Anthropic says it complied, but argues the government’s justification rests on a narrow potent

On Friday at 5:21 pm ET, the directive landed in Anthropic’s hands: shut off access to Claude Fable 5 and Claude Mythos 5 immediately.

Anthropic announced on X that it complied—but the company’s tone made clear it thinks the government misread the risk. The order disables both models for all users worldwide. not only the foreign nationals the export-control action was nominally aimed at. Access to Anthropic’s other models isn’t affected.

For Anthropic, this isn’t just another policy tweak. Mythos is described by the company as its most capable AI model. Anthropic previewed it in early April and kept it tightly restricted. arguing its exceptional ability to find security vulnerabilities in software was precisely what made broad release too risky. In Anthropic’s account, Mythos identified flaws in every major operating system and web browser it tested. Rather than open it to the public. Anthropic launched a controlled program called Project Glasswing. sharing the model with roughly 50 vetted organizations.

Those organizations included Amazon, Apple, Google, Microsoft, and CrowdStrike, according to Anthropic.

Claude Fable 5 was released just three days ago, and it was built to satisfy the commercial pressure that followed Mythos. Anthropic said Fable 5 is essentially a version of Mythos fitted with guardrails that block responses in high-risk areas such as cybersecurity and biology—guardrails the company argued make it safe enough for general release. It also positioned Fable 5 as the most capable AI model available to the public. citing benchmark tests from Vals AI. a company that tracks AI performance.

The government’s directive is framed as an export control action, restricting foreign national access. But in a lengthy blog post. Anthropic said its understanding is that the underlying concern is a claimed jailbreak of Fable 5. So far, the company says, the government has provided only verbal evidence of a “potential narrow, non-universal jailbreak.”.

Anthropic describes what that alleged jailbreak would involve: prompting the model to read a specific codebase and identify software flaws. The company adds that this “level of capability” is already widely available in other publicly accessible models. including OpenAI’s GPT-5.5. It also says the same approach is used routinely by cybersecurity professionals for defensive purposes.

Anthropic’s counterargument goes deeper than the jailbreak itself. The company says its strongest safeguards operate through independent classifier systems that function separately from the model. In other words. even if someone could convince Fable 5 to keep talking past a refusal. Anthropic argues the protections against the most dangerous outputs would still be in place.

None of that, apparently, stopped the government from acting.

Anthropic didn’t hide its frustration. “We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. ” the company wrote. It added: “If this standard was applied across the industry. we believe it would essentially halt all new model deployments for all frontier model providers.”.

The timing matters, too. Anthropic is widely expected to pursue an IPO this year. and it has built much of its public identity around presenting itself as the safety-conscious alternative to rivals. The irony. as observers see it. is that Anthropic’s own caution—its insistence that Mythos was too dangerous to release publicly because of its vulnerability-finding ability—has now apparently drawn precisely the kind of scrutiny that can disrupt a fast-moving business.

OpenAI’s Sam Altman may not feel sympathetic. In April. Altman told podcaster Ashlee Vance that Anthropic’s handling of Mythos amounted to “fear-based marketing.” He said it was like selling the world a bomb shelter: “It is clearly incredible marketing to say. ‘We have built a bomb. We were about to drop it on your head. We will sell you a bomb shelter for $100 million,’” Altman said.

Altman didn’t predict a government shutdown. But in the same conversation, he identified the problem that has now come back to bite Anthropic: when a company spends months telling the world its AI is uniquely dangerous, the world—including the U.S. government—tends to listen.

Anthropic Claude Fable 5 Claude Mythos 5 U.S. government export control national security Project Glasswing AI safety cybersecurity vulnerabilities IPO