UK AI Security Institute Regains Access to US AI Models After 17-Day Suspension

The UK’s AI Security Institute (AISI) has successfully regained access to pre-release versions of advanced AI models from leading American developers, including Google’s Gemini 4 Argon. This access was restored on October 11, 2026, following a 17-day period during which the institute's ability to evaluate these models was suspended due to a directive from the US government.

The suspension began on September 24, 2026, when the Office of the National Cyber Director requested that US safety reviews take precedence over international evaluations. This led to a situation where companies like Anthropic chose to withhold their models, such as the Claude Mythos 5.1, from non-US entities.

UK AI Minister Kanishka Narayan played a key role in advocating for the restoration of access, emphasizing the importance of maintaining strong relationships with developers. AISI officials have reiterated their commitment to collaboration with AI companies, focusing on the shared goal of enhancing cybersecurity through advanced AI.

AISI was established in 2023 following the international AI Safety Summit at Bletchley Park, with the mission to evaluate cutting-edge AI systems before they are released to the public. The institute employs over 100 technical experts dedicated to safety evaluations and has previously assessed models like OpenAI’s GPT-6 Astra, contributing to global standards for AI assessment.

The restoration of access is significant for both AI developers and governments, as it alleviates concerns about fragmented safety testing across national lines. This development may also influence other countries looking to establish collaborative AI regulations. As AISI continues its evaluations, the industry will be watching closely to see if more developers resume sharing their models with international evaluators.