Andrew Bailey, the Governor of the Bank of England, has warned that rapid advances in artificial intelligence could destabilise the global financial system. The warning was delivered in a two-page letter addressed to the G20 network of finance ministers and central bank governors. Bailey issued the letter primarily in his capacity as chair of the Financial Stability Board, the international body responsible for monitoring risks to the global financial system. The letter identifies so-called frontier AI models, the most advanced tier of the technology currently in development, as a source of systemic risk. It does not name a specific model or company. It does not set a timeline for when the risk might materialise.

This is not a warning about job losses or misinformation.

It is a warning that the machinery of global finance may be absorbing a technology faster than anyone can predict what it will do.

In October 2008, the collapse of Lehman Brothers demonstrated how a single point of failure, hidden within a densely interconnected system, could propagate losses across continents within days. Regulators at the time had modelled credit risk extensively. They had not modelled the speed at which automated trading and interlinked balance sheets could transmit a shock. Bailey's letter draws an implicit parallel. Frontier AI systems are being integrated into trading, risk assessment and lending decisions at a pace that outstrips the industry's understanding of how they behave under stress.

The mechanism concerns autonomy. Older software followed fixed rules that engineers could audit line by line. Frontier models increasingly make decisions through processes their own developers cannot fully trace, and they are gaining the ability to act on those decisions with limited human oversight. When such a system is embedded in a bank's trading infrastructure or a clearing house's risk engine, an error or an unexpected interaction with another automated system could move markets before a human notices.

Bailey's letter states plainly that this integration, if it continues to outpace regulation, could result in a global economic downturn.

The claim is significant because it comes from the institution charged with watching for exactly this kind of risk. The Financial Stability Board was established after the 2008 crisis for that purpose. Its chair naming AI as a threat on the scale of the risks that body was created to prevent marks a shift in how central bankers are framing the technology, from a productivity tool to a potential source of contagion.

Bailey joins a growing list of senior figures who have raised similar concerns in recent years, including technologists, economists and regulators who argue that the pace of AI deployment has outstripped the institutions meant to govern it. What distinguishes this intervention is its source. Central bank governors do not typically speak in the language of existential risk. They speak in the language of stability, capital buffers and stress tests. Bailey's letter uses that language to describe a technology that did not exist in its current form five years ago.

The letter is expected to feed into discussions at the next round of G20 finance meetings, where AI oversight is likely to feature alongside more familiar items such as sovereign debt and trade tensions. The Financial Stability Board is anticipated to consult member regulators on whether existing frameworks for operational risk are adequate for AI-driven systems, or whether a dedicated regime is required. No firm proposals have yet been published.

What Bailey has done is put a name to a risk that regulators have discussed informally for years.

The next question is whether the G20 acts on it before the system it protects is tested.

If history is any guide, the answer will arrive only after something breaks.

‍