An article summarized by Politico:
The White House has asked OpenAI and Anthropic to give the U.S. government first access to their newest AI models for security testing before providing them to the U.K.’s AI Security Institute (AISI). The request reportedly came from the Office of the National Cyber Director and reflects the administration’s position that American AI systems should be reviewed by U.S. officials before being shared with international partners. Anthropic appears to have followed the request, limiting early access to its latest Claude model to U.S. organizations.
The move could complicate the U.K.’s longstanding relationship with major American AI companies. AISI, established in 2023, has routinely received early access to frontier models and has become an important part of the U.K.’s approach to AI safety testing. U.K. officials have emphasized that AI risks cross national borders and said they intend to continue working closely with the U.S. and leading AI companies, even as Washington pushes for greater U.S. government access before international sharing.
The White House’s request comes as concerns grow over the capabilities and security risks of increasingly powerful AI models. U.S. officials are seeking additional resources for the Commerce Department’s Center for AI Standards and Innovation (CAISI), which is expected to play a larger role in testing advanced systems but currently has limited staffing. OpenAI has also called for CAISI to work with international AI-safety organizations, including the U.K.’s AISI, on global standards for monitoring AI development and safety.
