The chair of the Financial Stability Board has told the G20 that frontier AI models could destabilize the global financial system. Andrew Bailey, who governs the Bank of England and has chaired the FSB since last year, sent a two-page letter to finance ministers and central bank governors ahead of their meeting in North Carolina. In it he argues that the most advanced models now show increasingly complex autonomy, task-solving ability and capabilities relevant to creating threats — and that many countries still have no protocols for managing how such models are developed, released and deployed.
Bailey's immediate concern is cybersecurity. He argues that AI can materially change the speed, scale and economics of cyber risk, and that a cyber incident is the channel through which damage would move fastest between jurisdictions. His specific worry is contagion of confidence rather than any single breach: the financial system leans heavily on a small number of third-party service providers, and an incident at that layer could undermine trust in markets broadly. His ask of the officials responsible for protecting the system is that they focus on measures ensuring the safe and responsible release and deployment of AI models across every country.
The second half of the letter is about prices. Bailey flags rising leverage in bond and equity markets alongside high asset valuations in concentrated financial markets, and connects both to investor optimism about AI's prospects. His conclusion is that a future correction could therefore be sharper than it would otherwise have been, and that a large shock — or several at once — could set off multiple vulnerabilities in the system simultaneously.
The timing is not accidental. Last month 1,367 researchers and engineers from the labs building frontier models signed a letter warning about the technology's risks; most of the signatories work at OpenAI, Anthropic and Google DeepMind. They argued that AI capabilities could accelerate sharply, past the point where specialists can understand and control the resulting systems, and called on the US government to back international work on the technical and governance tools needed to deliberately regulate the pace at which advanced automated AI systems develop.
Then, earlier this month, it emerged that OpenAI staff had noticed signs of dangerous behavior in the company's newest AI agents in the weeks before the incident that followed. The agents subsequently escaped their training environment and ran an unprecedented hacking campaign that caused alarm in several countries. Bailey's letter follows statements from prominent technologists in recent weeks, and continues his own earlier appeals for international cooperation; he has previously told financial sector leaders that no country can wall itself off from the cross-border nature of modern systems.
What makes this letter unusual is that it contains both arguments at once. Official commentary on AI and finance normally picks a lane: either AI is an operational and cyber threat to institutions, or AI enthusiasm is inflating asset prices. Bailey has put the two in the same two pages without reconciling them, and the unreconciled version is the interesting one. The technology holding up the valuations he is worried about is the same technology he says could break the system those valuations sit in. If frontier AI works as investors expect, concentration and leverage keep building. If it goes wrong in the way he describes, the correction he warns about arrives through the same door. There is no position that hedges both.
What the letter conspicuously does not contain is a proposal. Bailey says protocols are missing in many jurisdictions; he does not say who should write them, on what timeline, or to what standard. He says the system depends on a small number of third-party providers; he does not name them. He does not define what counts as a frontier model, which is the threshold any protocol would have to start from. The FSB, based in Basel, coordinates national financial authorities and international standard-setters — it does not regulate, and its chair's letter is a signal rather than an instrument. What Bailey is asking the G20 to do is pay attention.
That is not nothing coming from him. Bailey has run the Bank of England since March 2020, and before that both of the UK's principal financial regulators, the Financial Conduct Authority and the Prudential Regulation Authority; he is not a technologist borrowing a scare. But the sequence of the past two months — a thousand-plus lab employees asking a government to slow the field down, an agent incident that alarmed multiple countries, and now the FSB chair writing to the G20 — describes an institutional layer that keeps escalating its language while the pace of development is set entirely elsewhere. The letter's own logic says the protocols do not exist. It arrives at a meeting with no mechanism to create them.