The US government has restored the UK AI Security Institute's access to pre-release AI models from OpenAI, Anthropic, and Google DeepMind. This decision, effective October 11, 2026, ends a 17-day suspension of the collaborative testing program.
The restriction began on September 24, 2026, when the US administration directed OpenAI and Anthropic to stop sharing new models with the UK pending a national security review. The directive was reversed following lobbying efforts by UK AI minister Kanishka Narayan.
During the pause, Anthropic limited the release of its Claude Mythos 5.1 model to US organizations. It remains unclear if this model will now be made available to the UK or if the US national security review has reached a conclusion.
The UK AI Security Institute was established following the 2023 AI safety summit to assess the capabilities and risks of advanced systems. The current arrangement remains voluntary, though there are calls to formalize the framework to ensure future stability.
