White House Restores UK AI Safety Institute's Access to Frontier AI Models

The White House has reversed an earlier decision to block the United Kingdom's AI Security Institute (AISI) from pre-release testing on leading American AI models.

Under the revised stance, the UK evaluation agency will once again receive early access to advanced frontier models built by US-based AI developers, including OpenAI, Google DeepMind, and Anthropic. The institute is tasked with analyzing and probing AI systems for critical safety and security risks before they are released commercially to the public.

Reversing the September Restrictions

The initial blockade took effect in September, when the Trump administration instructed major US artificial intelligence developers to halt sharing their models with the British institute.

Following that order, AI developer Anthropic moved to restrict the UK AISI's access to its Claude Mythos 5.1 model. The disruption created a significant hurdle for the UK's safety testing framework, which depends on early access to assess risks such as cyber capabilities, weaponization, and systemic vulnerabilities.

The reversal follows direct diplomatic engagement led by UK AI Minister Kanishka Narayan, who pressured US counterparts to reconsider the shutdown of safety cooperation.

Diverging Approaches to AI Governance

The brief ban highlighted growing friction between London and Washington over the trajectory of artificial intelligence regulation:

• International oversight push: UK Prime Minister Andy Burnham used an address at the United Nations General Assembly to advocate for formalized international AI governance.
• US resistance: At the same UN event, former President Donald Trump warned that "globalist" organizations should not interfere with American AI development.
• National security warnings: UK intelligence leadership—including the heads of MI5, MI6, and GCHQ—has explicitly warned government ministers that emerging AI capabilities risk empowering hostile foreign actors and terrorist networks.

Stricter Reporting Rules Ahead

With access now restored, AI Minister Narayan is preparing an upcoming address to unveil new UK domestic safety initiatives.

The planned framework includes tighter reporting mandates for AI-related security incidents that impact critical national infrastructure, formalizing how both private developers and public agencies must document and mitigate vulnerabilities in deployed models.