UK Reverses US Restrictions on Pre-Release OpenAI and Anthropic Model Testing

The UK reached an agreement with the US allowing Britain's AI Security Institute to evaluate frontier artificial intelligence models from OpenAI and Anthropic before they are released, reversing a previous White House policy requiring prior US approval. The policy reversal follows earlier US directives where the Trump administration instructed American AI developers to withhold advanced models until US authorities deemed them secure, leading Anthropic to withhold its Mythos model from UK evaluators without explanation. UK AI Minister Kanishka Narayan announced a government cybersecurity task force alongside the pact, noting that two government pilot programs using frontier AI identified more than 1,000 vulnerabilities across state systems. Narayan acknowledged the inherent fragility of the arrangement, warning that nations dependent on foreign technology must remain clear-eyed regarding the permanence of cross-border access agreements. "When you’re reliant on someone else’s technology, you have to be clear-eyed about the permanence of any outcome." — Kanishka Narayan