UK Regains Pre-Release Access to US Frontier AI Models
By @sharedot · · 7 pages
- AI
- AI Policy
- UK AI Security Institute
Britain's AI Security Institute will again test leading US models before release, reversing a Trump-era rule that put White House approval first.
What Happened
The UK will once more be able to evaluate leading artificial intelligence models before they are released, reversing a US policy that had required White House approval first, UK AI Minister Kanishka Narayan announced on Monday.
According to The Telegraph, as relayed by ET Enterprise AI, US developers such as OpenAI and Anthropic can now grant the British government early access to their models without violating restrictions the Trump administration had imposed.
Narayan said Britain had worked closely with the US government to retain access for the country's AI Security Institute to its most powerful models ahead of release.
The Telegraph, which first reported the change, wrote that Narayan lobbied for the reversal, and NDTV Profit reports he credited a recent summit with King Charles III for facilitating conversations between ministers and tech executives.
Why It Reverses Course
The change is a sharp pivot after what Bloomberg reported last month: that Donald Trump's administration had told OpenAI and Anthropic to withhold their models until US authorities deemed them secure first, and earlier this year had blocked Anthropic from releasing its latest models to foreign nationals.
ET Enterprise AI reports the shift signals a softer stance on international AI safety cooperation, giving US developers greater flexibility to work with British authorities on assessing the risks of increasingly powerful systems.
It also follows Anthropic's decision, per Bloomberg's earlier reporting via NDTV Profit, not to share the latest version of its powerful Mythos model with the AISI — and not to explain why.
The reversal is a win for Prime Minister Andy Burnham, who has pitched the UK as 'honest brokers' between nations policing the technology.
The Institute Doing the Testing
Britain's AI Security Institute is the best-funded government agency focused on regularly evaluating powerful AI models from OpenAI, Anthropic and others, according to NDTV Profit.
Set up in 2023 following the world's first AI Safety Summit, as ET Enterprise AI notes, it forms a central part of Britain's attempt to keep a foothold in AI developments while playing a distinct political role from Brussels, which has aggressively regulated tech.
But the institute itself sounded a caution on Monday: in an official statement carried by MLex, it warned that testing models to ensure they are safe before public release is no longer enough, because recent incidents show inadequate controls could cause real harm even before a model is released.
Security Stakes and Rogue Attacks
Narayan framed the access in security terms, saying in a London speech, 'I want Britain to become one of the hardest countries in the world to attack using artificial intelligence,' NDTV Profit reports.
Mounting concerns about cybersecurity risks, including a series of rogue hacks of outside databases, have spurred international worries about the security of the most capable models, and some British politicians have pressured Burnham to adopt stricter regulation of American AI developers.
Narayan also announced a task force to protect the UK government from rogue attacks, saying two government pilots had used frontier AI to identify more than 1,000 potential vulnerabilities and reduce patching time.
He said he would work with other ministries 'to find weaknesses within government systems before attackers exploit them.'
What Comes Next
The durability of the arrangement remains an open question — even its beneficiary says so.
'When you're reliant on someone else's technology, you have to be clear-eyed about the permanence of any outcome,' Narayan told Bloomberg per NDTV Profit, though he added he was 'very confident' Britain will be able to test the latest generation of models.
The AISI, meanwhile, said in its Monday statement, as MLex reports, that evaluation practices are struggling to keep pace and current efforts cannot assure whether increasingly autonomous AI agents would behave as intended; while not seeking to mandate requirements for frontier developers, it wants to collaborate on voluntary practices and urges stronger oversight throughout model development.
Over the weekend, Microsoft president Brad Smith told the BBC, as NDTV Profit notes, that the UK is the best nation to lead on global regulation of the technology.
Sources
- ndtvprofit.com › UK To Evaluate AI Models Before Release, Reversing Trump Policy
- enterpriseai.economictimes.indiatimes.com › Trump administration allows US AI developers to share models with Britain for safety testing
- mlex.com › Testing AI models before release no longer enough, UK AI Security Institute warns