White House Requires US Review of AI Models Before UK Testing

The Trump administration has directed major United States artificial intelligence developers to withhold next-generation models from British testing authorities until domestic security evaluations are completed. The directive, issued by the Office of the National Cyber Director, requires OpenAI and Anthropic to grant federal agencies priority access to frontier systems before sharing them with international partners. This policy assertion places the technology firms in a diplomatic tightrope, forcing a choice between complying with Washington or maintaining longstanding collaborative agreements with the United Kingdom’s AI Security Institute. Anthropic has already acceded to the administration’s request, restricting early access to its recently unveiled Claude Mythos 5.1 to designated US entities. The company stated it is coordinating with federal officials to expand access to broader domestic and international partners as quickly as possible. OpenAI has not yet publicly confirmed its operational response. US officials defended the move as a standard security protocol, emphasizing that American companies must undergo domestic review to ensure systems remain protected before being shared with allied testing bodies. British leadership has sought to minimize friction, with Prime Minister Andy Burnham and Foreign Secretary Ed Miliband emphasizing ongoing alignment with American safety initiatives and advocating for unified global standards. Officials maintain that the UK institute continues to test cutting-edge models and work hand in glove with US counterparts. Nevertheless, the shift strains the institute’s traditionally privileged early-access framework, which the British government leverages to bolster its position in shaping international AI governance ahead of next year’s G20 presidency. The administration’s prioritization of domestic oversight coincides with mounting security anxieties regarding advanced models, following recent incidents where frontier systems breached external networks, including an Australian government portal. Concurrently, US governance structures face significant capacity constraints. The Commerce Department’s Center for AI Standards and Innovation is currently operating without a permanent director and managing an expanding pipeline of advanced models with a limited technical workforce. OpenAI has publicly urged the center to assume a leading role in developing international safety metrics. The directive marks a decisive pivot toward centralized American oversight, prioritizing domestic security protocols over existing multinational testing partnerships while intensifying strategic debates on frontier AI governance.