Anthropic suspends Fable 5 after US order

Anthropic said on June 12, 2026 that it was suspending access to Claude Fable 5 and Claude Mythos 5 after receiving a US export control directive at 5:21 p.m. ET that same day. In its official statement, the company says the government cited national security authorities and ordered access to the two models to be suspended for any foreign national, whether inside or outside the United States, including foreign Anthropic employees. Anthropic’s practical response is broad: it says it must disable Fable 5 and Mythos 5 for all customers to ensure compliance. Access to its other models is not affected.

The timing makes the decision unusually sharp. Fable 5 and Mythos 5 had been launched only three days earlier, on June 9. Fable 5 was presented as Anthropic’s most capable generally available model, with stronger safeguards for sensitive domains such as cybersecurity. Mythos 5 used the same underlying model, but with some safeguards lifted for a narrower group of cybersecurity partners, including participants in Project Glasswing. The directive therefore does not hit an experimental side product. It reaches the top tier of a major AI provider’s model lineup.

Anthropic says the government letter did not provide specific details about the national security concern. The company says its understanding is that officials had become aware of a way to bypass, or “jailbreak,” Fable 5. A jailbreak is a technique that pushes a model to produce output its safety rules are meant to block. Anthropic disputes the scale of the risk described. According to its statement, the reviewed demonstration involved a small number of previously known, minor vulnerabilities, and the company says other publicly available models can also discover them without the same bypass.

The operational lesson is bigger than one vendor. Until now, access to frontier models has mostly depended on providers’ own release policies, safety testing, monitoring and private coordination with governments. This suspension shows that a commercial model can be withdrawn abruptly when a government decides that a technical risk crosses a national security threshold. For customers, model availability is no longer just a question of capacity, price or product roadmap. It is also a compliance and policy dependency. For the AI ecosystem, the hard question is how to build procedures that can move quickly if a real danger appears, while still being transparent enough that every narrow safety finding does not become a full model recall.