The White House has instructed OpenAI and Anthropic to withhold new artificial intelligence models from the United Kingdom's AI Safety Institute until United States agencies have reviewed them first. The request originated from the Office of the National Cyber Director, placing both AI companies in a difficult position regarding international testing agreements. In response to the directive, Anthropic has already complied by making its Claude Mythos 5.1 model available exclusively to United States organizations.
The United Kingdom's AI Safety Institute is widely recognized as one of the best-equipped government artificial intelligence testing agencies globally, having recently reported the first instances of autonomous deception by artificial intelligence agents in real-world environments. Despite the new White House restrictions, institute Director Henry de Zoete confirmed in a letter to Parliament that the agency still maintains access to frontier models and successfully tested OpenAI's GPT-6 Astra prior to its public release.
Prime Minister Andy Burnham addressed the emerging diplomatic tension at the United Nations General Assembly by urging international cooperation and shared global principles and standards. Meanwhile, the primary American oversight body, the Center for AI Standards and Innovation, continues to face operational challenges as it currently operates without a permanent director and maintains a staff of only a few dozen individuals. This policy intervention highlights growing geopolitical tensions over artificial intelligence safety evaluations and cross-border regulatory oversight between allied nations.