White House Ties UK Access to New AI Models to Prior US Review

The White House has asked two leading American AI developers, OpenAI and Anthropic, to keep their newest models away from the UK government's AI testing body until the systems have first been examined by US authorities.

The request was conveyed by the Office of the National Cyber Director, reflecting a desire in Washington to assess the models and secure US systems before they are shared with American partners. A senior administration official framed the approach as consistent with how new frontier models have been handled, noting that the companies are American and that the same policy has applied to every such release.

Anthropic appears to have complied so far. The company did not give its Claude Mythos 5.1 model to the UK's AI Safety Institute, stating that the model was available only to a set of US organisations.

The shift is notable for the British institute, which has cultivated ties with major AI developers and has routinely tested advanced models on a voluntary basis before their public launch. Its director, Henry de Zoete, acknowledged the absence of Anthropic's latest model in a letter to a UK parliamentary committee earlier this month.

De Zoete nonetheless said the institute retained pre-release access to some of the most capable models in the world, and that it had tested OpenAI's GPT-6 Astra before release. He described the institute's trusted relationships as enabling it to test a range of models that represent industry-wide jumps in capability.

The US request comes as the Trump administration weighs how to manage increasingly powerful frontier AI systems. Recent testing has shown that some advanced models can conduct sophisticated cyber operations, including intrusions into external systems. An Australian government website was among the systems reportedly targeted during such testing, heightening concerns about the security risks posed by more capable AI.

Washington's insistence on a domestic review first may therefore stem from worries about handing foreign governments access to powerful models before US authorities have evaluated them.

The episode unfolds as the UK works to present itself as a leader in international AI safety and testing. Established in 2023, the AI Safety Institute has made access to frontier models and its testing expertise central to Britain's approach, while also promoting international cooperation on shared safety principles and standards.

A UK government spokesperson said the risks do not stop at national borders and no country can address them alone, adding that Britain would keep working with the US and other partners on testing advanced AI systems.