White House Gives UK AISI Access to Frontier AI

The White House has reversed a policy that previously barred the UK AI Security Institute (AISI) from conducting early safety assessments on leading American artificial intelligence models. Following diplomatic intervention by UK AI Minister Kanishka Narayan, the US government lifted restrictions imposed in September, enabling developers including OpenAI, Google DeepMind, and Anthropic to resume sharing pre-release software with British safety researchers. The initial restriction had immediately disrupted safety testing, leading Anthropic to block AISI access to its Claude Mythos 5.1 model.

This policy reversal concludes a period of geopolitical friction regarding AI governance. During a UN General Assembly meeting, UK Prime Minister Andy Burnham campaigned for international regulation of AI systems, a stance that drew a sharp response from Donald Trump, who publicly warned "globalist" entities against meddling in American AI innovation. At the same time, domestic security officials in Britain underscored the urgency of rigorous evaluations, with the directors of intelligence services MI5, MI6, and GCHQ advising government ministers that hostile actors and terrorists are actively leveraging advanced AI capabilities.

For AI safety practitioners and engineering teams, the restored pipeline allows third-party evaluators to audit frontier models for critical vulnerabilities prior to public deployment. The evaluation window provides red-teamers the opportunity to identify unexpected emergent behaviors and potential dual-use risks in frontier architectures before commercial rollout.

Looking ahead, the UK government intends to strengthen its internal safety framework. Minister Narayan plans to unveil a series of new AI safety measures during an upcoming address. Among these proposals are stricter mandatory reporting rules requiring tech companies to disclose AI-related security breaches that affect critical infrastructure, ensuring that government assessors maintain visibility over severe national security threats.