Anthropic's Claude Fable 5: White House demands jailbreak fixes

Anthropic must address identified vulnerabilities before the company can rerelease Claude Fable 5, Trump administration officials say. The model was taken offline with export controls last week over concerns about jailbreaking, and the National Security Agency concluded there are ways to disable Fable 5’s guardrails.

What is the dispute?

The White House wants Anthropic to remediate jailbreaks in Claude Fable 5 before any rerelease, and officials have positioned the problem as the company’s responsibility to fix. Officials told reporters that Anthropic reiterated its position that the jailbreaks are minimal in a technical meeting on Monday with the Commerce Department and the Office of the National Cyber Director, Sean Cairncross, but the NSA concluded there are methods to disable the model’s safeguards.

The safeguards at issue block access to capabilities tied to the Mythos model, specifically capabilities related to cybersecurity, chemistry, and biology. Commerce’s Center for AI Standards and Innovation and the NSA have told officials they lack the staff or bandwidth to hunt down every potential jailbreak across all frontier models, and the administration expects Anthropic to test continuously and escalate findings to government channels.

Can Anthropic stop jailbreaking?

No definitive fix exists today, according to the reporting: independent cybersecurity experts increasingly view model guardrails as a temporary barrier rather than a permanent solution, because skilled users and future AI models will find ways to bypass constraints. The administration insists Anthropic must be more proactive about testing not just Fable 5 but all of its frontier models and flag potential jailbreaks to government teams, but security specialists doubt that any company can guarantee total prevention.

Anthropic has maintained that the administration’s concerns are overblown and that the practical effects of the jailbreaks are minimal. Nonetheless, three people familiar with the discussions said the government now considers it Anthropic’s problem to resolve, rather than the agencies’ task to police each vulnerability themselves.

Why it matters

If the NSA’s finding that Fable 5 guardrails can be disabled stands, the dispute frames a larger regulatory question: who is responsible for ongoing security testing of powerful models, and how much can technical safeguards realistically contain misuse. The Commerce Department’s Center for AI Standards and Innovation and the NSA signaling limits on their capacity shifts pressure onto vendors to build testing regimes and disclosure pathways, changing the compliance burden for companies deploying frontier systems.

This matter also ties export controls to operational security. Claude Fable 5 was taken offline with export controls last week; that step made a market-access decision into a national-security enforcement action and set a precedent for requiring demonstrable mitigation before a model returns to service.

What to watch

Watch whether Anthropic agrees to an ongoing public or government-facing program of internal red teaming and reporting, and whether it seeks to document mitigations that would satisfy Commerce and national security officials. The immediate signal will be any concrete remediation steps Anthropic announces, and whether regulators accept them as sufficient for a rerelease of Fable 5.

A White House spokesperson declined to comment on the matter.