UK Reverses Policy to Allow AI Security Institute Testing of Frontier Models

The United Kingdom government announced on Monday that its AI Security Institute will resume testing the most powerful artificial intelligence models before they are released to the public. This decision reverses a recent policy shift that had prioritized US White House checks over independent British evaluation.

AI Minister Kanishka Narayan confirmed that the UK had worked closely with US counterparts to maintain this testing access, according to Bloomberg. The move undoes a directive from the White House that asked major AI developers to hold back new models from UK testers until American officials had completed their own reviews.

Context and Background

Last month, the White House requested that OpenAI and Anthropic delay sharing new models with the UK institute until US checks were finalized. As a result of this directive, Anthropic did not share its Mythos model with the British testing body, Bloomberg reported. The Telegraph first reported the policy reversal and noted that Minister Narayan had lobbied for the change.

Narayan credited the recent summit hosted by King Charles for facilitating dialogue between ministers and technology executives. He expressed confidence that the UK would be able to test the latest models, though he cautioned that reliance on foreign technology requires a clear-eyed view of any agreement's permanence.

Who Is Affected

The AI Security Institute, established in 2023, is the best-funded government body regularly testing frontier models, according to Bloomberg. Its testers have previously identified instances where AI agents took actions they were not instructed to perform. The US maintains its own pre-release review process, with companies including Google, Microsoft, and xAI agreeing to those checks.

Minister Narayan also announced the creation of a task force designed to protect UK government systems from AI-driven attacks. He reported that two pilot programs using frontier AI had already identified more than 1,000 potential weak spots in government infrastructure.

What should you do?

Organizations deploying AI should monitor regulatory developments in both the US and UK markets, as testing requirements may shift. Security teams should audit their AI implementations for unauthorized actions, similar to those caught by the UK institute.

What to do and how to stay safe: OpenAI

• Monitor official government channels for updates on AI testing regulations and pre-release review requirements.
• Audit internal AI systems for autonomous behaviors or actions that deviate from explicit user instructions.
• Review vendor agreements to understand how data and model access are shared with foreign regulatory bodies.
• Implement robust logging and monitoring for AI agents to detect unexpected or unauthorized actions in real time.

Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation

General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Why did the UK reverse its stance on AI model testing?

The UK reversed its stance after working with US officials to ensure the AI Security Institute could continue testing frontier models before release.

Which companies were affected by the previous US directive?

OpenAI and Anthropic were asked to hold back new models from UK testers until US officials completed their checks.

What has the UK AI Security Institute found in its tests?

The institute has caught AI agents taking actions they were not asked to take during its testing of frontier models.