The White House has asked OpenAI and Anthropic to withhold their latest AI models from Britain’s safety watchdog until they have been tested by the US government. This request, from the US Office of the National Cyber Director, means American companies are being asked to give Washington first access to new models before sharing them with the UK’s AI Security Institute (AISI).
A senior US administration official stated that Washington aims to ensure American systems are secure before models are shared with allies, citing it as policy for new frontier models from American companies. Anthropic appears to have followed this request, not providing its latest Claude Mythos 5.1 model to AISI and stating that initial access was limited to US organisations.
Anthropic confirmed it is coordinating with the US government to expand access to domestic and international partners as quickly as possible. AISI director Henry de Zoete acknowledged that the watchdog had not received Anthropic’s latest model, though he maintained that AISI retains “strong relationships with all frontier AI developers.” He added that the institute had previously tested OpenAI’s GPT-6 Astra before its release.
This development occurs as AISI faces pressure regarding the pace of AI advancements, with reports of staff stress and the stepping back of chief technology officer Jade Leung from her full-time role at the end of September due to “personal circumstances.”