The UK can once again examine cutting-edge AI models before they hit the market, after a US policy that put the White House in the gatekeeper seat was rolled back. Announcing the shift on Monday, UK AI Minister Kanishka Narayan said Britain has been working with the US government "to retain access for Britain's AI security institute to the most powerful AI models before they are released." In London, he framed the mission bluntly: "I want Britain to become one of the hardest countries in the world to attack using artificial intelligence."
Bloomberg wrote last month that Donald Trump's administration instructed OpenAI and Anthropic PBC to delay new systems until US officials approved their safety. The UK's newly restored access was first revealed by the Telegraph, which said Narayan pushed for the change. According to him, meeting with King Charles III at a recent summit created new channels with ministers and technology chiefs, which in turn helped clinch the deal.
The UK's AI Security Institute, set up in 2023, is the best funded government shop dedicated to regularly testing advanced models from OpenAI, Anthropic and others. It sits at the core of Britain's plan to stay close to AI breakthroughs and to carve out a political approach distinct from Brussels, which has taken a hard line on tech regulation.
The US has lately tightened the reins on who can access Silicon Valley's most capable systems. Earlier this year, Trump's government barred Anthropic from making its latest models available to foreign nationals. Bloomberg also reported that Anthropic withheld the newest version of its Mythos model from the AI Security Institute and offered no explanation. While several prominent tech leaders have argued for shared global rules, Trump has dismissed that idea.
Over the weekend, Microsoft president Brad Smith said to the BBC that the UK is best placed to lead global AI regulation.
Who gets to test a model before release shapes what counts as safe. Market Briefs covers AI policy free every morning.
Concerns about cyber risk have been piling up, including reports of rogue hacks targeting outside databases, and UK lawmakers have pressed Prime Minister Andy Burnham to tighten oversight of American AI firms. Against that backdrop, Narayan announced a new task force focused on shielding the UK government from hostile activity. He pointed out that two government pilots used frontier AI and uncovered more than 1,000 potential vulnerabilities, and that efforts to speed up patching are underway.
Narayan added that he will coordinate with other ministries to assess AI capabilities and to spot weak points in government systems before attackers do.
Early, independent testing gives the UK a shot at catching flaws before new models power products, headlines and hype cycles. Fewer surprise security blowups can reduce operational risk for companies building or buying AI tools in Britain. Narayan sounded realistic but upbeat about how long any US access deal will last: "When you're reliant on someone else's technology, you have to be clear-eyed about the permanence of any outcome." Still, he said he's "very confident that, as a consequence of the conversations we've had, we will be able to test the latest generation of models."
Pre-release access is quietly becoming the core regulatory lever. Get the free Market Briefs daily newsletter and follow it.
