Washington Ends Its 17-Day Freeze on UK Model Testing

Britain's safety testers can once again get their hands on frontier AI models before the public does. On October 11 the White House lifted a freeze that had, for 17 days, blocked the UK's AI Security Institute from running pre-release checks on unpublished models from OpenAI, Anthropic and Google DeepMind.

The pause began on September 24, when the Office of the National Cyber Director told the labs to hold new models back from British testers until Washington had finished its own national security review. For a little over two weeks, one of the few government bodies that evaluates frontier systems before release was left waiting on the outside.

Why the freeze mattered

Pre-release access is the whole point of a safety institute. The value of an outside evaluation lies in seeing a model before it ships, when there is still time to flag a capability that should not go out unguarded. Cut off that early access and the institute is reduced to testing what everyone else already has.

The episode also stranded specific models. Anthropic confirmed it had held back Claude Mythos 5.1, its restricted-access model, during the freeze. "We're coordinating with the US government to expand access to a broader set of domestic and international partners as quickly as possible," the company said.

An American-first rationale

According to CNBC's reporting on the review, the administration's position came down to a blunt principle: these are American companies, and American reviewers go first. The reversal followed lobbying by the UK's AI minister, Kanishka Narayan, and ended with British access restored to the three labs' pipelines.

Accounts differ on the exact scope. Some reports describe the freeze as covering only OpenAI and Anthropic, others include Google DeepMind, and at least one notes that individual companies had set their own terms for international testers. What is clear is that the pause is over and the pre-release channel has reopened.

Whether it stays open is another question. President Trump has been vocal in his opposition to curbs on what he calls "super intelligence," a stance that sits awkwardly alongside a review process built to slow models down before release. The same tension runs through the voluntary White House accord on super intelligence signed earlier this month, and through the federal safety inquiry into OpenAI and Anthropic now underway. For now, the testers are back at the bench.