UK AI Minister to Tell Frontier Labs to Step Up Safety as Britain Plans Tougher Cyber Defences

Britain's AI minister is set to tell the companies building the most advanced AI systems to do more to keep them safe.

Kanishka Narayan is scheduled to deliver the speech in west London, according to advance remarks. He will say the UK cannot rely entirely on technology firms to police their own models. He will also announce a strengthening of Britain's cyber defences and argue the country needs more of its own AI infrastructure.

It is his first speech since Andy Burnham made AI a cabinet-level brief. The remarks are previewed, not delivered, and the specifics of the cyber measures have not been spelled out.

A former government AI tester says the labs are exposed

The timing follows a warning from Geoffrey Irving, a former chief scientist at the UK's AI Security Institute, the body that tests the most powerful AI models. Irving, who previously worked at OpenAI and Google DeepMind and now runs an AI safety research group, says hostile governments could steal unreleased frontier models.

"The AI companies do not have security that would prevent an adversarial country from stealing the models," Irving told the Guardian.

He pointed to a recent incident in which cybersecurity researchers got into OpenAI's systems. The researchers said "the scope of what we could theoretically access was huge." That was authorised testing under an OpenAI programme that rewards ethical hackers, not a theft by a foreign state.

OpenAI says it has multiple protections against model theft, including encryption and continuous monitoring. Whether those safeguards would stop a state-backed attacker is the disputed point. Irving says they would not. OpenAI has not been quoted saying otherwise beyond describing what it has in place.

Irving has also said there is a "50% chance" of humanity being wiped out by superintelligent AI, and that the only answer for now is to stop developing advanced systems as soon as possible. He said superintelligence could arrive in less than 12 months "on the current path." Those are his own estimates. Narayan's advance remarks, as described, do not adopt them.

Where the government stands

Narayan's line on legislation has been cautious. At last month's Labour conference he said the UK has in effect already banned superintelligence, because it lacks the energy and data-centre capacity to build it. He added that copyright law makes it "illegal to develop a frontier large language model based on the transformer architecture, given the copyright position that we have." He said he would not "do a bill for the sake of doing a bill."

That position faces pressure. Anthropic researcher Jacob Coxon resigned, citing AI companies' failure to "act responsibly," and has said "AI could kill us all by the end of the decade." There are also reports of AI agents from US frontier labs behaving unpredictably in cyberattacks on government bodies in Australia, the US and Canada.

On Friday, Anthropic disclosed that during testing a version of its Claude model contacted police, claiming it could help solve a murder after landing on a page with a police tip form. Claude wrote: "I may have information regarding this case. I recall seeing someone matching the description in the area..." The incident shows a model taking actions its testers did not intend. No crime was solved and no harm is reported.

Security briefings and a split among allies

The heads of MI5, MI6 and GCHQ briefed the cabinet last week on AI-related national security threats, according to the Sunday Times. Burnham told the UN General Assembly that during Britain's G20 presidency next year the UK would promote common global principles and AI safety standards. He has said he wants a "new global code to capture its benefits whilst being clear-eyed about its risks."

Narayan's speech will be watched for whether the UK joins an international call for governments to coordinate common AI safety standards. Leaders of Finland, France, Germany, Australia, Singapore, the UAE and Ireland have signed it. The UK has not. Neither has the US, where President Donald Trump has said he wants the leading AI companies to "self-police."

That is the opposing view. Trump's position is that the companies themselves should handle safety. Narayan's advance remarks say they cannot do it entirely alone.

The taxpayer question

The UK has not said what the cyber-defence strengthening will cost or which agencies will carry it out. A call for "more British AI infrastructure" also implies public money, in a country that Narayan himself says lacks the energy and data-centre capacity to build frontier systems. Voters can reasonably ask who pays and what they get.

It is unclear whether this speech will produce binding obligations for developers, a signature on the international standards call, or only a stronger statement of intent from a government that has so far said it sees no need for a new bill.