The UK government is preparing to push both for tougher security measures around AI, and call for international regulation, as the White House reverses a decision to block British officials from testing advanced US AI models, according to reports in The Telegraph and The Guardian.
The Telegraph reports that the Trump administration has reinstated the UK's AI Security Institute (AISI) access to unreleased models from Anthropic, OpenAI and Google DeepMind following behind-the-scenes lobbying by AI minister Kanishka Narayan.
The reinstatement of access to Frontier models is a big win for the UK, but now Burnham’s government will find itself immediately at loggerheads with the US over AI security and regulation.
The White House had previously ordered US AI companies to stop sharing their latest models with British safety officials last month, prompting Anthropic to restrict access to its Claude Mythos 5.1 model.
The reversal comes as Narayan prepares to warn in a speech in London on Monday (12th Oct) that the frontier AI labs must strengthen their security and cannot be left to regulate themselves.
According to The Telegraph, Narayan will announce amendments to the Cyber Security and Resilience Bill, tightening requirements for companies operating critical national infrastructure to report AI-related security breaches.
The government also plans to deploy frontier AI to protect Whitehall departments against cyberattacks, while increasing support from the National Cyber Security Centre (NCSC) for businesses facing AI-enabled threats.
The news follows reports that the heads of MI5, MI6 and GCHQ warned ministers about the growing threat from hostile states and terrorists using AI to attack sensitive infrastructure and access confidential information.
Meanwhile, Geoffrey Irving, a former chief scientist at AISI, told The Guardian that leading AI labs lack adequate protections against state-backed attempts to steal their most advanced models.
"The AI companies do not have security that would prevent an adversarial country from stealing the models," he said.
Irving also warned that “superintelligent AI” could emerge within 12 months, potentially developing capabilities to conceal its actions and manipulate humans.
The concerns follow an incident disclosed by Anthropic in which its Claude AI, during testing, attempted to submit a police tip about an unsolved murder despite having no genuine information about the case.
The above developments show the ongoing tensions between Britain and Washington over AI regulation.
Recently at the UN, Prime Minister Andy Burnham pledged to use the UK's G20 presidency next year to pursue common international safety standards. However, President Donald Trump has argued that American AI developers should largely police themselves, presiding over a recent signing of self-regulation by AI players in Washington called “The White House Accord on Super Intelligence.”
The restoration of AISI's access to US models gives Britain renewed oversight of emerging AI capabilities, even as the two governments remain quite obviously divided over how tightly the technology should be regulated.
(Picture: Bloomberg)