The instruction is gone barely a month after it was issued. According to reports, the White House has dropped its decision to bar the UK’s AI Security Institute (AISI) from getting early access to new AI models. The original order had told America’s leading AI labs to stop handing their unreleased models to British safety testers.
With the reversal, AISI goes back to its normal job of checking AI systems for safety risks before they are released. Models from Anthropic, Google DeepMind and OpenAI will again reach the agency ahead of launch.
How the block played out
The dispute started in September, when the Trump administration told the companies to stop sharing their models with AISI.
The effect was easy to see. After the directive arrived, Anthropic limited access to its new Claude Mythos 5.1 model. Pre-release testing only works if labs hand their systems over, and for a while at least one major developer did not.
London pushed back, and won this round
The U-turn came after diplomatic pressure from AI Minister Kanishka Narayan, who does not see the restored access as the end of the matter. In an upcoming speech, Narayan intends to announce new AI safety measures, including tougher reporting rules for AI-related security incidents that affect critical infrastructure.
The two governments are still not on the same page. Prime Minister Andy Burnham had earlier argued for international AI regulation at the UN General Assembly. At that same gathering, Trump warned that “globalist” organizations should keep out of American AI development.
Britain also has its own reasons to want AISI back in the process. The heads of MI6, MI5 and GCHQ had told ministers that AI was empowering hostile actors and terrorists. When all three of a country’s intelligence chiefs say that, losing early access to the most powerful models is a serious setback for its safety testers.
For the moment, access has been restored. But Washington shut it off once with a single instruction to the labs, and it could do so again.