Anthropic has abruptly disabled its newest models, Claude Fable 5 and Claude Mythos 5, for all users. The company says the move follows an emergency export control directive issued by the US government on June 12.
Fable 5 had been released for general use only days earlier, while Mythos 5 was limited to selected organizations and research partners. Both were derived from Mythos Preview, a high-capability system built for security research that several companies had used to find and fix hundreds of software vulnerabilities.
A letter from the Commerce Secretary
According to reports, Commerce Secretary Howard Lutnick sent a letter to Anthropic chief executive Dario Amodei stating that strict export restrictions were being placed on Mythos 5 and Fable 5. Under the order, access is barred not only for people outside the United States but also for foreign nationals inside the country.
Anthropic says the government is concerned that a party has found a way to bypass the models’ safety limits, a so-called jailbreak, in a manner that could create national security risks. The company’s position is that the technique in question exposed only a few minor, previously known security flaws, and that other publicly available AI models are capable of finding the same kind of weaknesses.
Anthropic calls it a likely misunderstanding
The company also says Fable 5 already shipped with multiple safeguards, and it has described the government’s decision as probably resting on a misunderstanding. Anthropic said it would release more details within 24 hours and would work with officials to restore access to both models.
What changes for users and developers
Following the shutdown, new chats from Claude users are being routed automatically to other models, mainly Claude Opus 4.8, while existing sessions running on Fable 5 are ending with errors. Developers using the API have been told to migrate their applications to other models.
The episode is the first instance of the US government intervening this directly and quickly against a frontier AI model. It is likely to open a new debate over AI security, national security, export law and the shape of future AI regulation.
