Britain’s AI Security Institute has regained early access to frontier models from Anthropic, Google DeepMind and OpenAI after the White House reversed a reported restriction. The institute evaluates systems for safety risks before public deployment.
The US administration had told companies in September to stop sharing new models with the British agency, according to reporting cited by The Decoder. Anthropic subsequently limited the institute’s access to Claude Mythos 5.1. Diplomatic pressure from UK AI Minister Kanishka Narayan preceded the change in policy.
Restored access lets British evaluators examine current capabilities rather than relying only on models that have already shipped. Narayan is also expected to announce tighter reporting requirements for AI-related security incidents affecting critical infrastructure.
The reversal repairs one testing channel but does not create a binding international safety regime. Laboratories still control what they provide, evaluators need enough time and technical access to run meaningful tests, and governments may disagree over publication or deployment decisions. The immediate result is practical: one of the best-established public model-testing teams can again receive systems from three leading US developers before release.