Reports emerged that AI company Anthropic failed to submit its latest model, Claude Mythos 5.1, to the UK's AI Security Institute (AISI) for safety testing, reportedly following pressure from President Trump. The move has alarmed British politicians across the political spectrum, who warn it could leave the UK unable to assess and manage risks from cutting-edge AI systems, with some describing it as a threat to national security comparable to, or greater than, that posed by hostile states.

The development follows Trump's order in June for Anthropic to restrict foreign nationals' access to its cyber security-focused models over national security concerns, prompting the firm to say it would disable its most advanced models for non-US users. It also comes amid wider unease about AI safety, after an Anthropic researcher resigned this week accusing AI firms of a "hubristic gamble" with humanity, and a colleague suggested AI carried more than a 10 per cent chance of causing human extinction within a decade. Figures including Reform UK's Danny Kruger and shadow AI minister Ben Spencer called on the government to secure continued access to frontier models, while Prime Minister Keir Starmer acknowledged AI's security risks when questioned in Parliament by Lib Dem leader Ed Davey.