The White House has reversed a previous directive that blocked the UK’s AI Security Institute (AISI) from evaluating pre-release AI models from major US developers. Following diplomatic intervention by UK AI Minister Kanishka Narayan, the AISI will regain early safety testing access to upcoming models from Anthropic, Google DeepMind, and OpenAI, according to a report by The Daily Telegraph.
The initial block was implemented in September under the Trump administration, prompting Anthropic to restrict access to its Claude Mythos 5.1 model. The policy shift comes as British security officials from MI6, MI5, and GCHQ warned that unchecked AI capabilities could empower hostile actors and terrorists, reinforcing the need for pre-deployment safety evaluations.
Following the access restoration, Narayan plans to introduce stricter security reporting requirements for AI incidents involving critical national infrastructure. The decision highlights ongoing tension between international AI governance initiatives and nationalist US policy stances regarding domestic technological development.
Why it matters
Frontier AI startups must navigate evolving bilateral safety evaluation frameworks when deploying models internationally.
Pre-release safety auditing access remains intact for UK regulators, signaling continued reliance on cross-border government testing.
Enterprise developers should anticipate mandatory security reporting requirements for AI deployments tied to critical infrastructure.
Source: the-decoder.com



