White House seeks US review before UK tests frontier models

The White House reportedly asked OpenAI and Anthropic to hold new frontier AI models back from the UK’s AI Security Institute until U.S. officials reviewed them first. Reports published September 24, 2026 identified the Office of the National Cyber Director as the office behind the request, framing model access as a cybersecurity and national security concern.

Anthropic’s Claude Mythos 5.1 is the only model clearly named in the reporting. Anthropic said the model was “only available to a set of U.S. organizations,” and did not provide it to the UK institute, which runs prerelease evaluations for risks including cyberattack assistance, biological weapons uplift, and loss of control.

The scope of the White House request remains unclear. A senior U.S. administration official described the approach as policy for every new frontier model from American companies, but no public statement has confirmed whether it is a formal directive or a one-time request. The UK institute’s director, Henry de Zoete, told Parliament it still has prerelease access to some of the world’s most capable models, suggesting the dispute is limited to specific releases rather than a broad cutoff.