The most powerful people in global finance spent the last weekend of August in Asheville, North Carolina. Waiting for them was a two-page letter, and its warning was blunt: the newest, most capable AI systems have become a risk to the stability of the financial system itself.

The letter came from Andrew Bailey — not only the Governor of the Bank of England, but the current chair of the Financial Stability Board (FSB), the international body that coordinates financial regulation for the G20. He sent it on 28 August, ahead of a meeting of G20 finance and central-bank deputies held on 29–30 August, and it was published on 31 August.

What Bailey actually said

Bailey's central claim is that "frontier" AI models — the largest, most general systems at the leading edge of the field — are "showing increasingly sophisticated autonomy and problem-solving abilities, as well as threat capabilities."

The part he singled out as most urgent was not a market bubble or job losses. It was cyber-security. The impact of AI on cyber risk, Bailey said, is the most immediate concern for the global financial system, because the technology changes the speed, scale and economics of an attack. In plain terms: a capable model can help a smaller, less-skilled attacker do more damage, faster and for less money than before — and a cyber-incident at one institution can spread across borders before regulators have caught up.

His second point was about preparedness. Many countries, Bailey warned, do not have the systems in place — the protocols and oversight — to manage how advanced AI models are developed, released and deployed. The risk is global; the rulebooks are national and patchy.

Why a central banker is talking about AI at all

It can seem odd for the person who sets UK interest rates to be writing to the world's finance ministers about AI safety. But the FSB's job is precisely to spot risks that cut across the whole system rather than sitting inside one bank or one country — and a technology that reshapes both financial markets and the tools available to attackers lands squarely in that remit.

There is also a hard-nosed reason banks are paying attention. The financial sector is one of the earliest and heaviest adopters of AI, from fraud detection to trading to customer service. The same capabilities that make the technology useful to a bank make it useful to whoever is trying to break into that bank — and the defenders and attackers are now drawing on the same rapidly improving toolkit.

This is not the first warning — and that is the point

Bailey has been building this case for months, and it is worth separating the strands so they do not blur together.

In July, the Bank of England's twice-yearly Financial Stability Report put a number on the market side of the risk: it estimated that a sharp correction in AI-related shares could knock around 2.2 percentage points off UK output — even though none of the big AI companies are listed in London, because UK pension funds and savers hold heavy exposure to US equities through index-tracking funds.

Separately, the FSB has been working on the supervisory response. Its consultation on "sound practices for the responsible adoption of AI" opened in June, with a final report expected in October. And in Europe, the European Systemic Risk Board issued its own warning over the summer about systemic cyber risks stemming from frontier models.

The G20 letter is the moment those separate threads were pulled together and put in front of the people who actually set financial policy across the largest economies.

What it means

For now this is a warning, not a rule. Bailey is not proposing a specific law; he is telling finance ministers that the oversight gap is real and that coordination has to happen internationally, because a national rulebook cannot contain a cross-border risk.

The honest reading is that regulators have noticed the gap faster than they can close it. The FSB's own timeline — a consultation in June, a final report in October — moves at the pace of institutions, while the models it is worried about are updated in weeks. That mismatch is the uncomfortable core of Bailey's letter: the people responsible for financial stability are now openly saying that the technology is moving faster than the systems meant to govern it.

Whether the G20 acts on that, or files it alongside the other warnings, is the thing worth watching between now and the autumn.