AI and Regulation
Frontier AI models are named a cyber risk to the financial system
By Staff Writer | 1 September 2026
The Financial Stability Board's chair has told G20 finance ministers that many jurisdictions have no protocols for releasing advanced models, and that the consequences of a failure will not stop at a border.
The Financial Stability Board published its chair's periodic letter to G20 finance ministers and central bank governors on 31 August 2026, ahead of their meeting in Asheville, North Carolina. The letter is dated 28 August and is signed by Andrew Bailey, who chairs the board and is Governor of the Bank of England. Its second section is given over to frontier artificial intelligence, and that is what makes it worth reading rather than filing.
The board describes frontier models as showing increasingly sophisticated autonomy and problem solving abilities, and threat capabilities alongside them. It then states plainly where the danger to the financial system sits.
For the financial system, the most immediate concern is the potential impact of frontier AI on cyber risk.
Andrew Bailey, Chair of the Financial Stability Board and Governor of the Bank of England
Speed, scale and the cost of an attack
The letter's argument is about economics rather than novelty. Frontier models may alter the speed, the scale and the cost of a cyber attack, and the board's view is that a change in those three at once could undermine confidence across the system, because the technology providers the system depends on are so few. Firms and authorities are told to expect a higher volume of vulnerabilities and a faster pace of patching, and warned that the pace itself creates a problem if testing and recovery cannot keep up with it safely.
What is asked of institutions is specific. They should be able to restore critical systems and data from what the letter calls bare metal after a serious incident, and they should prepare for scenarios in which several firms, or several users of one shared technology, are disrupted at the same time.
The letter records that many jurisdictions have no protocols governing the development, release and deployment of advanced frontier models, and puts fixing that first.
Borrowed money and concentrated bets
The first section is about markets. The board reports a rise in the use of borrowed money in equity markets, through geared exchange traded funds and momentum strategies, including by private investors, and a growing presence of geared funds that are also exposed to sovereign debt. Its concern is not the borrowing on its own but the way it meets high valuations and a concentrated market, and in particular the growing cross investment between artificial intelligence companies and the operators of very large data centres. Asset valuations tied to artificial intelligence are described as stretched.
Why a construction reader should care
Two reasons, and neither is about banking. The first is that the same concentration argument applies to any sector that has moved its records, its models and its programme controls onto a handful of platforms. A project bank, a common data environment and a design authoring platform are third party dependencies of exactly the kind the letter describes.
The second is the pipeline. If valuations tied to artificial intelligence are stretched and the cross investment between model builders and data centre operators is what worries the board, then the schemes being designed and priced on the strength of that investment carry a risk that is now recorded in a G20 document rather than in a broker's note.
The board coordinates national financial authorities across 24 countries and jurisdictions and reaches about 70 more through its regional groups. It has not proposed a rule here. It has written down where it is looking, which is usually the year before anything is required of anyone.