The United States has restored the UK AI Security Institute's access to unreleased models from American labs, the Telegraph reported on October 11, as summarized by Startup Fortune. The hold lasted 17 days. It began on September 24, when the White House Office of the National Cyber Director told AI companies to withhold new models from British testers until Washington finished a national security review.
The hold covered pre-release sharing with the institute by Anthropic, OpenAI, and Google DeepMind. It was an instruction to companies. No rule was proposed and no statute changed. During the freeze Anthropic kept Claude Mythos 5.1 with a set of US organizations, and UK officials did not review it. The reversal followed lobbying by UK AI minister Kanishka Narayan and other ministers.
Two things are still unknown. The outcome of the US security review has not been reported, and the White House has issued no public statement on the reversal. The stated reason for the hold was security: the administration wanted American systems locked down before sensitive capabilities reached any foreign partner, allies included.
The episode exposed how the UK's arrangement works. Britain has no AI-specific statute. The institute, founded in November 2023 and renamed in February 2025, tests frontier models under access agreements negotiated with the labs. An agreement of that kind can be paused by a letter from another government, and for 17 days it was.
Bottom Line
Access is back, and the arrangement that was interrupted is still voluntary on every side. If your product depends on a third-party evaluation before launch, find out whether that evaluation rests on a contract or on goodwill. If you sell models into the UK, watch whether ministers move toward legislation now that they have seen the gap.