The White House has reversed its previous decision to block the UK AI Safety Institute from accessing advanced artificial intelligence models. Following diplomatic pressure, the agency will once again receive early access to frontier systems developed by major labs including Anthropic, Google DeepMind, and OpenAI to test for safety risks before deployment.
The restriction originated in September when the Trump administration instructed AI companies to halt sharing their models with the British safety institute, prompting Anthropic to restrict access to its new Claude Mythos 5.1 model. The block was lifted after intervention from UK AI Minister Kanishka Narayan, who is preparing to announce stricter reporting requirements for security incidents involving critical infrastructure.
This reversal comes amid broader international debates over artificial intelligence governance and security. While UK leadership has advocated for collaborative global regulation, security officials have raised concerns about AI empowering hostile actors, making the institute's safety evaluations a critical component of international oversight.