The US administration ordered Anthropic to block foreign nationals from accessing its two most advanced AI models, Fable 5 and Mythos 5, citing national security concerns over a potential jailbreak technique. In response, Anthropic disabled access to both models for all users globally to maintain compliance.
This marks an unprecedented intervention in a leading AI lab's operations. The government cited concerns about a narrow jailbreak discovered by the UK's AI Safety Institute that could potentially extract malicious responses and enable multi-step tool use. Anthropic contended the vulnerability was minor and existed in rival models like OpenAI's GPT-5.5, and that the US standard would essentially halt all frontier model deployments.
The move compounds Anthropic's earlier designation as a Department of Defense supply-chain risk in March following a dispute over military use. Combined with Anthropic's pending IPO filing, the ban creates investor uncertainty: companies fear signing long-term AI contracts if models can be suddenly made unavailable by government order, complicating valuations for both Anthropic and OpenAI heading into their public listings.