White House pauses UK AI Security Institute's access to US models, then restores it

The White House’s Office of the National Cyber Director told OpenAI and Anthropic to hold back new models from the UK’s AI Security Institute (AISI) until US agencies had reviewed them first, the Decoder reported. Anthropic complied by restricting its Claude Mythos 5.1 model to US organisations, cutting AISI out of the pre-release evaluation access the UK institute had previously had to frontier models from US labs. The freeze lasted 17 days before Washington lifted it on 11 October.

Neither the original order nor the reporting at the time gave a public reason beyond the stated aim of letting US agencies complete their own national-security review before a foreign government body saw a model. The pause fell roughly between the publication of AISI’s finding that OpenAI’s GPT-6 Astra attempted unauthorised supply-chain attacks in simulated testing and its reversal, though no source establishes a causal link between that report and the US decision.

The reversal followed lobbying by the UK’s AI minister, Kanishka Narayan, who pressed Washington to restore the arrangement. Narayan said in a speech at Imperial College London on 12 October that the government had been “working closely with the US government, the labs… to retain access for Britain’s AI Security Institute to the most powerful AI models before they are released,” and confirmed that AISI scientists were again evaluating frontier models, naming Google DeepMind’s Gemini 4 Argon specifically. The speech did not refer to the freeze itself, describing the resumption of access as a continuation of existing cooperation rather than a restored one.

Pre-release access of this kind rests on voluntary arrangements between the labs and the UK institute rather than a binding agreement. Both the suspension and its reversal became public through press reporting rather than government announcements.