Why US has barred foreigners from accessing Anthropic’s Mythos and Fable AI models

The U.S. government issued an export control order on June 13 blocking all foreign access to Anthropic’s AI models Fable 5 and Mythos 5, including for foreign employees. The order, with unspecified national security concerns, follows reports of a jailbreaking method targeting Fable 5, though Anthropic claims the vulnerabilities are minor and already known in other models like OpenAI’s GPT-5.5. The company stated the government provided only verbal evidence of a narrow jailbreak technique involving codebase analysis, which it argues is widely replicable. Anthropic’s blog post emphasized that such capabilities are standard among defenders securing systems, undermining the urgency of the restrictions. Fable 5 and Mythos 5, launched this week, are Anthropic’s most advanced publicly available models, built on its Mythos-class architecture. The company had previously restricted its even more powerful Mythos Preview model due to fears of misuse, including its ability to expose vulnerabilities in major operating systems and web browsers. Mythos Preview was discussed for limited access in India but never publicly released. The export controls follow earlier Pentagon concerns about Anthropic’s AI in military operations, where the company resisted use for autonomous weapons or domestic surveillance. The restrictions come as Anthropic prepares for a U.S. public listing, with a valuation of $965 billion. The models include safeguards to block sensitive queries, rerouting high-risk requests to less capable systems.