White House tells AI giants to hold models back from UK safety watchdog

The White House has instructed AI companies OpenAI and Anthropic to withhold their latest models from the UK's AI Security Institute (AISI) until they have been reviewed by the US government. This request threatens the UK's privileged early access to frontier technology. The US Office of the National Cyber Director is asking the American companies to provide Washington with the first look at new models before sharing them with the UK's AI Security Institute.

A senior US administration official explained that the US wants to ensure their systems are secure before models are shared with allies. Anthropic appears to have already complied, not providing its latest Claude Mythos 5.1 model to AISI, stating access was initially limited to US organizations. AISI director Henry de Zoete acknowledged that the watchdog had not received Anthropic’s latest model, but said it still maintains strong relationships with all frontier AI developers.

AISI has previously tested OpenAI’s GPT-6 Astra before its release. The restrictions come as AISI faces increasing pressure over the rapid advancements in AI, with staff reportedly taking time off due to stress and concerns over increasingly capable systems. The move also comes amid calls for international cooperation on AI safety, with AI executives such as OpenAI's Sam Altman and Anthropic's Dario Amodei advocating for global safety standards at the UN.

However, the US administration has rejected new global AI governance, creating a divide over the control of the world's most powerful models. This could complicate Britain's G20 presidency next year, where AI safety was to be a focus under Andy Burnham's leadership.