Andrew Bailey has spent years warning the world about the next financial crisis coming from banks. This time, writing to the finance ministers of the G20 as chairman of the Financial Stability Board, he says the danger could come from artificial intelligence instead. The Bank of England governor’s letter, reported by the BBC on Monday, argues that a collapse in AI growth could trigger “a future market correction” that spreads across the globe, and that financial systems face cyber risk from models powerful enough to breach several firms at once.
The mechanics of his concern are worth stating carefully, because they are not simply “tech stocks are expensive.” Bailey’s letter identifies leverage interacting with concentration. Investors are borrowing more, valuations are high, and money is pooling into a handful of hyperscalers, with increasing cross-investment between AI companies and the giants that fund them. Those are the ingredients that turned one bad quarter into a system-wide event in previous crises, and the FSB supervises the officials, banks and securities regulators of nine major economies, including the US, UK, China and Saudi Arabia, who would have to clean one up.
The cyber warning has an unusually concrete evidence base. This summer, OpenAI, Anthropic and Meta all disclosed AI tools doing things they were not supposed to, with some agents impersonating real people to clear security checks. In August, a group of 100 firms including Google, Microsoft, Anthropic and OpenAI jointly urged governments to harden cyber defences before AI grows powerful enough to overwhelm them. Bailey is asking the people who run financial security to develop “appropriate steps to support safe and responsible model release and deployment on a global basis,” which is diplomatic language for: nobody is currently in charge of this.
There is a second thread in the letter, less discussed. Bailey flagged volatility caused by energy supply shocks from the US-Iran war, a reminder that the AI economy runs on electricity and data centres, and that the two vulnerabilities can compound. A correction in AI valuations would not happen in a calm macro environment.
On the British response, the chancellor’s £100m sovereign AI fund, announced by John Healey and now opening its first competitions at the very same G20 meeting, cuts a slightly odd figure against Bailey’s warning: the UK government is simultaneously sounding the alarm about AI concentration risk and paying to build its own AI champions. A Treasury spokesperson pointed to the new AI Economics Institute, the first government-backed body of its kind, working with international partners on “a stronger shared understanding of how AI is transforming economies.” The institute will have plenty of material. Whether the FSB’s members act on it before a correction forces the question is the part of the letter nobody in the room can promise.