The UK’s AI Security Institute (AISI) has regained the ability to evaluate advanced AI models from major US developers, including OpenAI and Anthropic, before their public release. This restoration of pre-release access follows a reversal by the White House on October 11, 2026, reinstating a voluntary arrangement that had been paused just weeks earlier.
Policy Shifts and Their Impact
The pause, initiated by a White House directive in September 2026, required US developers to withhold new models from AISI until US reviews were completed. This sequencing rule, which granted American officials a first look at sensitive AI capabilities, had tangible consequences. Notably, Anthropic restricted the release of its Claude Mythos 5.1 model to US organizations, adhering to the White House request.
The October 11 reversal effectively unwinds this restriction, allowing US developers to once again share pre-release models with AISI under the voluntary framework that predated the September directive. This means British testers can examine models for safety risks and potential misuse prior to their public launch.
UK’s AI Safety Ambitions
This development did not occur in isolation. UK AI Minister Kanishka Narayan actively advocated for British evaluators to regain access. His case reportedly emphasized the established trust and early access that underpinned UK relationships with developers like OpenAI, arguing that the pause undercut these foundations.
The AISI, established following the AI Safety Summit at Bletchley Park in November 2023, focuses on evaluating advanced AI systems for cybersecurity and biological threat risks, feeding its findings into UK AI governance policy. The UK is also considering proposals for mandatory incident reporting, which could introduce a compulsory layer to current voluntary testing arrangements.
Developer and Geopolitical Considerations
For AI developers, the policy reversal alleviates a potential logistical challenge. Anthropic’s decision regarding Claude Mythos 5.1 illustrated how government directives can influence product availability. A return to the previous arrangement reduces the likelihood of fragmented, country-specific product launches.
Washington’s recent focus has been on domestic security, prioritizing US review of sensitive capabilities. London, meanwhile, aims to solidify its position as an independent and trusted evaluator for companies and governments.
Navigating Conditional Cooperation
However, the voluntary nature of the access arrangement means it remains subject to developer goodwill and potential future shifts in White House policy. This situation highlights the conditional nature of international AI safety cooperation, which is susceptible to policy shifts and relies on voluntary agreements.
While the UK’s regained access is a positive step, it underscores the ongoing need for more stable frameworks for AI safety evaluation. The UK’s consideration of mandatory incident reporting, if adopted, could provide a more consistent channel for regulators, independent of developer cooperation or immediate White House approval.
Why This Matters
The materials describe a narrow update: The UK’s AI Security Institute (AISI) can again evaluate advanced AI models from US developers like OpenAI and Anthropic before their public release. Whether the voluntary pre-release access arrangement will continue in the future, as it depends on developer goodwill and White House preferences.
Broader Context
Source materials place the factual news in this context: The UK’s AI Security Institute (AISI) can once again evaluate advanced models from major US developers, including OpenAI and Anthropic, before those models reach the public.
