Andrew Bailey has sent the G20's finance ministers and central bank governors a two-page letter about artificial intelligence, in his capacity as chair of the Financial Stability Board rather than as governor of the Bank of England. The distinction matters: this is a financial stability document, not a technology one, and it contains two arguments that deserve to be separated because they are not the same kind of claim.
The first is that a collapse in AI-sector growth could produce a market correction that spreads. Bailey names the mechanism plainly: highly priced equity markets, increased borrowing by investors, and a growing concentration of money into a small number of large technology companies.
Read that list without the words "artificial intelligence" in front of it and it is the oldest description in finance. Elevated valuations, leverage against them, and concentration are the three ingredients of every credit event that has ever been written up afterwards. The asset class changes; the arithmetic does not.
The familiar half
That is not a criticism of the warning. It is the reason to take it seriously, because we know what this pattern does. Concentration is the part already visible in the published data — index concentration has already changed what a diversified portfolio contains without most holders choosing anything. Leverage is the part that is hardest to see until it is tested, which is the same problem private credit is discovering in Sydney, where forty lenders held forty views of one borrower and none of them had been priced.
What makes the AI version unusual is not the mechanism but the physical spending attached to it. When AI spending moved from experimentation to infrastructure, the exposure stopped being a set of positions and became buildings, substations and multi-year equipment orders. Those do not reprice quickly. A correction in the securities can happen in a week; the concrete takes a decade and gets finished regardless.
The second argument is the one worth reading twice, because it is genuinely new and it is not about prices at all. Bailey writes that frontier models are showing increasingly sophisticated autonomy and problem-solving abilities as well as threat capabilities, and that they may materially alter the speed, scale and economics of cyber risk. He then connects that to a highly interconnected financial system in which disruption can spread across jurisdictions.
That is a different claim from the first, and it does not resolve into a familiar shape. A market correction has a known transmission path and a body of policy built for it: capital buffers, liquidity requirements, resolution regimes, a lender of last resort. There is no equivalent apparatus for a system in which the cost of mounting a competent attack has fallen by an order of magnitude while the number of institutions worth attacking has not.
The economics point is the sharp end of it. Cyber defence has always relied on attacks being expensive to mount well — expertise is scarce, and scarcity rations how many serious attempts a system faces. A tool that makes competent attacks cheaper does not add a new risk to the register; it changes the rate at which every existing entry occurs.
Both halves of the letter are addressed to the same audience, and only one of them has an established playbook. Regulators know how to write rules about leverage against concentrated, highly priced assets, even when they cannot make anyone follow them in time. Nobody has a capital buffer for the second thing.
The contents of the letter, its length, its recipients and the quoted phrases are as reported by the BBC and other accounts of Andrew Bailey's two-page letter to G20 finance ministers and central bank governors in his capacity as chair of the Financial Stability Board, published 31 August 2026. The analysis is our own.
Topics marketsaiconcentration
