Governments should not be rushing to regulate artificial intelligence, the governor of the Bank of England has said.
Andrew Bailey, who holds the key oversight post in UK finance, stressed the need for “rigorous model testing, conducted both before and after deployment” ahead of any tighter regulation.
He recognised the potential benefits of Frontier AI were “immense” but needed “rigorous testing” to better understand its “increasingly complex behaviour”.
But experts have warned that AI used in Gaza and Iran already demonstrate its risks to humanity and say it should be regulated.
In an article accompanying the bank's Financial Policy Committee quarterly meeting report, Mr Bailey admitted the risks posed by sophisticated AI were “real and increasingly significant”.
“Models will behave unexpectedly,” he said. “They will expose weaknesses in our assumptions. That is not evidence that testing has failed, it is evidence of why testing is necessary.”
He said regulators needed to learn from the increasing number of incidents and near misses as AI systems are being developed but then establish “credible points of intervention”.
“One should not see testing as an alternative to future regulation, nor as a complete solution,” Mr Bailey said. “Over time, a more formal regulatory framework may well emerge. But regulation is not, in my view, the right place to start.”
Earlier this month, OpenAI and Anthropic reported their AI agents had gone rogue and hacked into other companies' systems, prompting further speculation about its threat to humanity.
After a meeting with AI executives on Tuesday, US President Donald Trump said they had agreed to establish voluntary standards and “independent auditors” to assess whether systems were working as designers intended. Leading tech companies also said they would work to ensure their AI tools do not “hack or access technical systems in unintended ways”.
Speaking at the White House, Meta owner Mark Zuckerberg said they agreed to develop “robust internal controls” and Mr Trump added he is considering a 10-member board to police the safety of AI tools.
Specialists in military AI have argued that waiting for a clearer understanding of the technology is inadequate, as the technology already presents existential threats for humans in Gaza and Iran, where it has been extensively used.
Existential already
But Dr Matt Mahmoudi, assistant professor in digital humanities at the University of Cambridge and a researcher and adviser on AI and human rights at Amnesty International, said current debates about AI human-extinction events ignore what is already happening.
“Human extinction level events distract the conversation away from existing existential threats that artificial intelligence is already posing to people elsewhere,” he said.
He highlighted the Israeli AI-assisted targeting systems, such as Lavender, Gospel and Who’s Daddy?, that are understood to have contributed to the huge death toll in Gaza.
The National has previously reported from witnesses in the territory that social media posts and other pictures were inputted into data, making targets of people who might have no connection to Hamas. Furthermore, at times Israeli operatives took only 20 seconds to verify a target before the person was struck by drone or air power.
“The harms are real, they are happening,” Dr Mahmoudi told a Labour conference fringe event on AI and autonomous weapons.
Legally binding international rules were needed to protect civilians from autonomous AI weapons, he said. But he warned current international humanitarian law could be difficult to enforce when it might be unclear whether it was a human or machine that was responsible for the kill decision.
Dr Alexi Drew, an AI specialist at technology consultancy Penumbra Analysis, said the rapid development of autonomous AI weapons meant attempts to regulate them were already struggling to catch up.
“We’re trying to create effective legal constraint to a thing which is already out in the wild,” she said, highlighting the use of increasingly autonomous technology in the Gaza and Ukraine wars.
“It’s already been released. It’s out there. It’s been fired from the metaphorical gun,” she said. “We can’t put it back in.”
That governments had missed the opportunity to prevent the technology’s development created a more pressing obligation to regulate its use.
“We have an even greater responsibility to legislate and constrain safely the thing that we have contributed to the creation of,” she said. “We missed the opportunity to stop it and therefore we now have a greater responsibility to ensure that the thing we created, the Frankenstein’s monster”, is properly controlled.
Governments are currently divided over what international rules to introduce for lethal autonomous weapons, with disagreements during UN talks over whether new legally binding controls are needed or current guidelines followed.
