Anthropic shut down Fable 5 and Mythos 5 after receiving a formal letter from the US Commerce Department at 5:21 p.m. on June 12. Signed by Commerce Secretary Howard Lutnick, the order required the company to immediately block all foreign nationals from accessing the two newly launched models. To comply, Anthropic cut off access not only for global customers but also for its own foreign employees.
The timeline was short. Fable 5 and Mythos 5 had been live for only three days after their June 9 release. According to Axios, citing a government official, the Commerce Department had already tried several times before launch to persuade Anthropic to delay the rollout because of concerns over the models’ strong cyber capability, but those efforts did not succeed.
Commercial and restricted versions were both pulled
Anthropic had used a split release structure. The public commercial version, Fable 5, came with strict filters and would hand off prompts involving cybersecurity, biology, or chemistry to the less capable Claude Opus 4.8. Mythos 5, the less restricted model, was limited to the “Project Glasswing” program and made available only to about 50 vetted organizations, including red-team groups, critical infrastructure operators, and defense units.
Anthropic said that in the first weeks of Glasswing, Mythos helped those partners uncover more than 10,000 high-severity and critical vulnerabilities in important software. The company also said testing showed the model could identify and exploit zero-day flaws across every major operating system and every major browser when instructed to do so.
A reported jailbreak became the trigger
The immediate catalyst was a claim from a third-party company, made three days after launch, that it had successfully jailbroken Mythos 5 in its controlled setting and reported the result. That gave the Commerce Department the basis to move quickly. Anthropic disputed that account and said its review found the demonstration had targeted the public Fable 5 filter rather than Mythos 5 itself.
According to Anthropic, the method involved asking the model to inspect a specific piece of code, locate software flaws, and suggest fixes. The result, it said, revealed only a small number of minor vulnerabilities that were already known. The company added that the technique was simple and could also be reproduced with other public models, including OpenAI’s GPT-5.5.
Export controls applied directly to an AI model
The Commerce Department used export-control authority that has traditionally been applied to items such as chips and weapons. Under the letter, any future export, re-export, or in-country transfer of Fable 5 or Mythos 5 will require a separate license application, and violations may bring civil and financial penalties. The restriction applies to all foreign nationals inside and outside the United States, including Anthropic employees. Other Anthropic models were not included.
Anthropic said in a statement that it viewed the move as a misunderstanding and was working to restore access as soon as possible. The company also argued that recalling a commercial model already deployed to hundreds of millions of users over a narrow potential jailbreak issue would create a standard that could halt new model rollouts across the sector.
One official said the two models may remain locked for the next several weeks while the government develops additional national security measures. A day before the shutdown, Anthropic CEO Dario Amodei had publicly argued that governments should have the power to block high-risk AI.