White House reportedly restores UK access to advanced AI models for safety testing

Britain will retain access to powerful AI models before they reach the public, AI minister Kanishka Narayan confirmed in London on October 12. The announcement follows a reported White House reversal of restrictions on sharing those systems with the UK AI Security Institute, restoring access for safety testing ahead of release.

September restrictions put US assessment first

In September, the Trump administration told companies to stop sharing new models with the institute, The Decoder reported. Anthropic subsequently restricted access to Claude Mythos 5.1. The instruction therefore had a concrete consequence for the British lab: access to that model was curtailed, rather than simply becoming a subject of diplomatic disagreement.

The reported restriction concerned the order of government scrutiny. AFP, citing media reports, said Washington had asked OpenAI and Anthropic not to share models with the UK lab before US authorities assessed them. That placed US assessment ahead of British access, rather than barring the UK from receiving the systems indefinitely.

The block was lifted after diplomatic pressure from Narayan, according to The Decoder, which cited the Daily Telegraph for the restored-access reporting. Narayan’s subsequent public statement confirmed that Britain had been working with the US government to preserve the institute’s access before models are released.

We have been working closely with the US government... to retain access for Britain's AI Security Institute to the most powerful AI models before they are released

Kanishka Narayan, UK AI minister, speaking in London on October 12

Restored access includes three leading developers

Anthropic, Google DeepMind and OpenAI will again provide early access, the Daily Telegraph reported, according to The Decoder. The institute, established in 2023, tests AI systems for safety risks before they ship. The restored arrangement concerns that pre-release testing role, with the British lab able to examine systems before public availability.

Narayan also identified a specific system already available to the institute: Google’s Gemini 4 Argon. He said the lab had been granted access to the model, which has not been publicly released over safety concerns, AFP reported. That disclosure describes access for government testing—not an announcement that the model is available to users.