The Office of the National Cyber Director has instructed OpenAI and Anthropic to prevent the U.K.'s AI Security Institute from accessing new frontier models until U.S. agencies finish their own security reviews. This mandate requires a specific sequence of events where the government evaluates the technology and hardens U.S. systems before distribution to international partners. Anthropic has reportedly complied by restricting its Claude Mythos 5.1 model to U.S. organizations, though OpenAI has not yet confirmed if it will follow the request.

A senior administration official stated that this protocol applies to every new frontier model because the developers are American companies. Washington aims to understand the capabilities of increasingly powerful AI and secure domestic infrastructure before granting early access to allies like Britain.