Anthropic cut off access to its newest AI models, Mythos 5 and Fable 5, late Friday. The shutdown came just days after their launch, driven by a sudden export-control order from the US Commerce Department. Under those rules, both models can’t be used outside the United States until further notice. All other Anthropic systems remain online. According to Anthropic’s statement, the company got only verbal evidence of a narrow jailbreak threat. Officials claim someone found a way around Fable 5’s safeguards to ask for cybersecurity, chemistry, or biology advice on a specific codebase. Anthropic says the only harmful results so far were reports of minor software vulnerabilities. It also notes that competing AI platforms—like GPT-5.5—offer similar probing capabilities. Axios sources add that the Biden administration wants time to bolster national security defenses against these kinds of AI-driven attacks. They’ve paused the models’ release until the “national security apparatus” can be hardened. That process might wrap up in a few weeks, according to one official.