US asks OpenAI and Anthropic to delay UK model tests until US review

The White House has asked OpenAI and Anthropic to withhold new AI models from the UK government's testing agency until US officials complete their own security review, according to a Politico report. The request, raised by the National Cyber Director, puts the companies in a difficult position, as the UK's AI Security Institute (AISI) has been a leading global tester with early access to models. Anthropic has so far complied, withholding its Claude Mythos 5.1 model from UK testers and limiting access to vetted US organizations. The UK's AISI, established in 2023, has sought to downplay any clash, while officials like Ed Miliband have called for rigorous global testing standards. In the US, the Commerce Department's CAISI faces resource constraints, though OpenAI has asked it to lead international AI safety standards work.