Bank of England Warns G20 Frontier AI Could Crash Markets

Andrew Bailey used a two-page letter to the G20 to argue that the newest generation of AI models has become a financial stability problem rather than merely an economic story. Writing as chair of the Financial Stability Board as well as governor of the Bank of England, Bailey sent the note ahead of the group’s meeting of finance ministers and central bank governors in Asheville, North Carolina. His framing is that frontier models now pair autonomy and problem-solving with something regulators have to price in directly: threat capability.

Cyber risk is what he calls the most immediate concern. The argument is not that AI creates a new category of attack but that it changes the arithmetic of existing ones — how fast an intrusion can move, how many targets it can reach, and how quickly confidence drains out of a market that suddenly cannot verify what it is looking at. Bailey stressed that this does not stay put geographically, because financial firms lean on the same handful of technology providers and shared infrastructure regardless of where they are regulated.

The second thread is valuation. Bailey said plainly that equity market valuations appear stretched for AI companies, drawing an explicit comparison to the peak of the dot-com era, and warned that leverage is now interacting with those valuations and with heavy concentration in a small number of names. Borrowing to buy equities is the mechanism that turns an ordinary repricing into something faster and more violent, and AI-linked assets are where that concentration currently sits.

What makes the letter unusual is the direction of the risk it identifies. Bailey noted that both material bottlenecks in AI progress and genuine conceptual breakthroughs could damage current valuations — a rare acknowledgement that a technology surprising to the upside can be as destabilising to a market priced on one set of assumptions as one that disappoints. Stack that on top of fragilities in private credit and sovereign debt markets, he argued, and the result is a correction that could turn disorderly and spread across borders.

None of this comes with a policy instrument attached, and that is worth being clear about. The FSB chair’s letter to the G20 is a signalling document; it flags what the board intends to examine and gives national regulators cover to start asking their own supervised institutions harder questions. The concrete effect, if there is one, tends to arrive months later in stress-test scenarios and operational-resilience rules rather than in anything a market can trade on the same week.

Still, the tone marks a shift. Central bankers have spent much of the past two years discussing AI as a productivity variable — something that might lift growth, might displace workers, might do neither on the timeline anyone expects. Bailey has moved it into the column where authorities keep the things that can break the plumbing, alongside leverage and sovereign debt, and that is a different conversation entirely.