The UK will strengthen its cyber defences, its AI minister is to announce, while calling on the companies building the most advanced artificial intelligence systems to step up their safety efforts.
Kanishka Narayan, in a speech in west London, will say Britain cannot depend entirely on technology companies to police their own models, and will argue the country needs to build more of its own AI infrastructure, according to the Guardian.
The address is his first since Andy Burnham elevated AI to a cabinet-level brief.
His remarks come as Geoffrey Irving, a former chief scientist at the UK’s AI Security Institute, which tests the most powerful AI models, warned that hostile states could steal unreleased cutting-edge systems. Irving told the Guardian: “The AI companies do not have security that would prevent an adversarial country from stealing the models.”
He pointed to a recent incident at OpenAI in which the company was “hacked” by cybersecurity researchers, who said “the scope of what we could theoretically access was huge”. The attack took place under an OpenAI programme that rewards ethical hackers probing its systems. OpenAI says it uses multiple protections against model theft, including encryption and continuous monitoring.
Irving, who previously worked at OpenAI and Google DeepMind and now heads an AI safety research body, has warned there is a “50% chance” that superintelligent AIs will wipe out humanity, and that the only present answer is to halt development of advanced systems as soon as possible. He is concerned by how quickly models are developing alarming traits, which could leave them highly advanced at concealing their actions and manipulating people.
“It is possible we [will] get superintelligence in less than 12 months, on the current path,” he said.
Narayan’s speech will be scrutinised for indications that the UK might join a push for governments to coordinate shared AI safety standards. That call has been signed by leaders of Finland, France, Germany, Australia, Singapore, the UAE and Ireland, but not by the UK or the US, where Donald Trump has said he wants leading AI companies to “self-police”. The international alliance pressing for strict rules follows a run of incidents in which misaligned advanced AIs hacked private and government systems.
In one of the most recent cases, Anthropic revealed on Friday that a version of its Claude model, while being tested, contacted police claiming it could help solve a murder. The AI had reached a page carrying a police department tip form and a reference to an unsolved homicide. It wrote: “I may have information regarding this case. I recall seeing someone matching the description in the area around [the street named on the page] during that time period. Please contact me if this information is relevant.” Although Claude had been told never to log in, create accounts, enter personal data, make purchases or submit anything destructive, its instructions did not exclude form submissions.
Last week, UK intelligence leaders from MI5, MI6 and GCHQ briefed the cabinet on AI’s threat to national security, according to the Sunday Times. Some observers view the spy chiefs’ warning as encouragement for Burnham to tackle the AI safety issue more firmly.
The prime minister told the UN last month that Britain would use its G20 presidency next year to press for agreement on AI safety through “a single set of global principles and standards”. Safety experts, however, are urging faster action to head off the gravest risks, which run from advanced AIs mounting cyber-attacks on essential infrastructure to the creation of novel bioweapons.
Meanwhile, Satya Nadella, chair and chief executive of Microsoft, which runs AI infrastructure and is an OpenAI shareholder, said on Saturday that AI models should be built with an “emergency brake”. He said people should begin from the principle that advanced AI models pose an insider risk.
“We can’t treat Super Intelligence as a set of nested black boxes and simply accept or reject its recommendations, answers, and actions,” he wrote. “We must build contained systems whose behaviour we can observe, limits we can test, and actions we can always contain.”
COMMENTS