World leaders must wise up to threat of cyber-attacks or rogue fridges will be least of their worries | Heather Stewart

Rogue French fridges were an unlikely talking point for the world’s leading economists, central bankers and finance chiefs meeting in a sultry Bangkok last week – helping to underline the spiralling cost of cyber-attacks.

Kristalina Georgieva, the managing director of the International Monetary Fund, seized on news that supermarket chains across France had to chuck out piles of spoiled food after hackers accessed their fridge controls and turned up the temperature to 50C.

Speaking at the fund’s annual meetings in the Thai capital, she was keen to hammer home the importance of guarding against “bad actors” in a rapidly changing, interconnected world – especially now that AI was “putting the potential of digital fraud on steroids”.

Policymakers have been warning for some time that the AI boom could at some point threaten financial stability via a share price crash – a “correction”, as it is euphemistically called.

But a different, related threat came into sharp focus in Bangkok. Georgieva said she was concerned that cybercrime, turbocharged by AI, could undermine trust in the financial system altogether.

Social media feeds are rife with adverts using celebrity fakes to punt cryptocurrencies or gold, and even the old-school email-style scams can be generated far more prolifically.

Just as a sea of AI slop has the potential to undermine trust in fact-based reporting and send people down political rabbit holes, Georgieva fretted that cyber-fraud could knock consumer confidence in mainstream banks.

Recalling her own childhood behind the iron curtain in Bulgaria, she said many people did not trust the system, and kept their cash in a shoebox under the bed.

“I worry, today, that unless we figure out how to get ahead of this curve … we may end up with lots of people with shoeboxes under their beds,” she said.

It might not even take a cybercrime wave. This week the JP Morgan chief executive, Jamie Dimon, warned that the bank would not allow digital agents to operate on its customers’ behalf. “You don’t want rogue agents, and there could be thousands of them,” he said.

AI agents, including Meta’s Muse, are starting to offer their users the opportunity to hand over their financial details. The agents will then scout for the highest savings rates and shift the user’s money accordingly.

Financial regulators fret about this for a number of reasons. One is that if such services become widely used, they could have destabilising herding effects. With deposits able to be moved around the financial system at a moment’s notice, it could accelerate a bank run in times of uncertainty.

A second worry is that highlighted by Dimon: who is responsible if your AI agent makes a financial decision on your behalf that turns out to be disastrous – or even criminal?

The answer is probably that we will need case law to find out, which will only come when something goes badly awry.

That is alarming because legal ambiguity about who owes what to whom when things go wrong can be an accelerant to financial instability.

We learned that lesson back in the 2008 financial crisis, when it turned out that some banks had built financial products with such fiendishly complex liability structures that they were effectively impossible to value – unleashing panic.

Speaking alongside Georgieva, the Bank of England governor, Andrew Bailey, who chairs the Financial Stability Board of global central bankers, pointed to another arresting news story from this week – that Anthropic had warned AI models might soon be able to “self-replicate” in domestic devices. Who should bear the responsibility – and the costs – if your internet-enabled oven, or telly, blows up as a result?

One entirely prudent response might be for consumers to reject internet-enabled devices and stick to “dumb” appliances instead that can’t be hijacked by self-replicating bots. This month, some of those French supermarkets rushed to take their fridges offline to avoid the hack.

While it may not be much of a loss to society if internet-enabled kettles and fridges do not take off, the financial analogue to a dumb appliance would be Georgieva’s shoeboxes – and that would have costs for those consumers, and the smooth-running of the economy as a whole.

As Bailey somewhat apocalyptically put it: “If we don’t have trust in money, we don’t operate as a society.”

Senior financial policymakers are genuinely disturbed at the collision between rapidly developing AI and the global financial system – which now extends well beyond heavily regulated banks – to embrace social media platforms, payments apps and other financial tools.

It is unfortunate this moment comes as the US – crucible of the AI boom – is governed by a man whose main contribution to the debate has been to seek to rename AI “superintelligence”.

At the Bangkok meetings this week, Kevin Warsh, the chair of the US Federal Reserve, obediently used Donald Trump’s phrase on stage – as all US government agencies have been ordered to.

There has been some progress in acknowledging the risks. Trump’s superintelligence taskforce, chaired by the director of national intelligence, Jay Clayton, is setting up mechanisms for AI labs to report “breaches, hacks, jailbreaks, and other ​issues” – although not, it seems, to submit their models for pre-launch testing.

The UK has pledged to take up the issue of AI safety as part of its agenda for the G20 presidency next year, and G7 finance ministers announced in Bangkok that they would carry out simulations to test the resilience of the global financial system to cyber-attacks.

That could help to highlight vulnerabilities, but the questions at stake are much wider and more complex than whether criminals can hack the banks – and are urgent now, just as the prospects for international cooperation seem bleak. Superintelligence will seem anything but super if its misuse becomes a trigger for financial chaos.