White House orders OpenAI and Anthropic to withhold their frontier models from the UK AI Safety Institute

Reliable reports revealed that the White House administration requested both OpenAI and Anthropic to refrain from granting the UK AI Safety Institute (AISI) access to their frontier models for testing and evaluation purposes, as reported. This request comes amid escalating disputes between the US and UK sides over AI governance mechanisms and information sharing.

Insiders familiar with the matter explained that the US move stems from growing concerns within the administration about sharing sensitive technical information related to frontier models with foreign entities, even if they are allies. These tensions reflect a shift in the US view of advanced AI models as strategic assets regarding which excessive transparency should be avoided.

In the same context, the United Kingdom established the AI Safety Institute as a globally leading body in evaluating the risks of frontier models, developing partnerships with major companies to secure early access for testing purposes. Critics argue that the US stance will impede efforts to build an effective international AI governance framework based on cooperation.

Policy experts warn that this direction might push Britain and other countries to develop independent evaluation capabilities away from US companies, potentially leading to fragmentation in global AI safety standards at a time when challenges demand more international coordination than ever. The issue is expected to spark an ongoing diplomatic debate between the two allies.