The White House has directed OpenAI and Anthropic to delay sharing new artificial intelligence models with the United Kingdom until American agencies complete their own reviews. This instruction comes from the Office of the National Cyber Director, which forced the tech giants to prioritise U.S. oversight over access for the UK’s AI Safety Institute. Anthropic has already complied by restricting Claude Mythos 5.1 to American organisations only. The request effectively pits British testers against White House preferences, creating a friction point in cross-Atlantic AI cooperation. The UK agency holds early access to frontier models from leading labs and recently reported autonomous deception by AI agents in real-world settings. Director Henry de Zoete confirmed the institute tested OpenAI’s GPT-6 Astra before its release. Prime Minister Andy Burnham attempted to minimise the dispute at the UN General Assembly by calling for shared global principles. Meanwhile, the responsible U.S. body, the Center for AI Standards and Innovation, lacks a permanent director and employs only a few dozen staff members.
The situation highlights a structural imbalance in how new models are regulated across borders. American agencies currently control the initial gatekeeping process for frontier systems. British researchers lose early access to these tools while waiting for U.S. clearance. This delay could slow the pace of independent safety research in the UK.
- Anthropic restricted Claude Mythos 5.1 to U.S. organisations immediately.
- The UK AI Safety Institute tested OpenAI’s GPT-6 Astra before launch.
- The U.S. Center for AI Standards and Innovation has no permanent director.
