Britain’s AI Security Institute is getting its access back. The White House scrapped its earlier move to cut the agency off from new models, and pre-deployment access from Anthropic, Google DeepMind and OpenAI will resume, according to The Daily Telegraph.
The freeze started in September, when the Trump administration instructed the labs to halt model sharing with the British testers. AI Minister Kanishka Narayan made the case through diplomatic channels, and his effort has since succeeded in lifting the block, while Anthropic had earlier curbed access to its new Claude Mythos 5.1 model.
Narayan seized the moment to roll out a broader risk package on Monday. At Imperial College London he announced that a frontier-AI cyber defense program, one that has already turned up more than 1,000 vulnerabilities across government systems, will be extended.
Behind the measures sit the Cabinet Office, AISI, the National Cyber Security Centre and the culture department. How serious AI incidents and near misses ought to be reported will come under review, along with how to answer agent failures that might unfold faster than conventional emergency planning allows for.
Microsoft president Brad Smith made the case that same day that Britain, rather than Washington or Brussels, is best positioned to lead on global rules. On a BBC podcast he described AI as both a weapon and a tool, adding that voluntary commitments bind only the companies already willing to comply.
The rules Smith proposes would include a legal duty to report breaches by rogue agents, outside evaluation ahead of release, and kill switches. His comments came as Trump warned against globalist bodies meddling in American AI work.
