UK regains US access to frontier AI models for safety testing

The United States has reversed a recent decision to restrict British access to unreleased frontier AI models. Following lobbying by UK AI Minister Kanishka Narayan, the UK’s AI Security Institute (AISI) has regained the ability to conduct safety testing on pre-release systems from major developers, including OpenAI, Anthropic, and Google DeepMind.

This restoration follows a 17-day pause initiated by the Trump administration, which had requested that developers withhold models pending a US national security review. The access is critical for the AISI to evaluate potential risks and misuse before technologies are made public.

Parallel to these developments, the UK government is moving to strengthen national cyber defences and AI resilience. Minister Narayan has called for increased safety efforts from AI developers, emphasizing that Britain cannot rely solely on the self-regulation of tech giants. The government is also exploring ways to build more autonomous AI infrastructure to mitigate the risks of technological dependency on foreign entities, particularly in critical sectors like healthcare, defence, and national infrastructure.