The Bank of England's Deputy Governor for Financial Stability told an ECB policy forum that the legal chain of accountability in finance — a human decision, a named person — breaks when autonomous agents execute trades and move payments at scale.
A senior Bank of England official told a European Central Bank policy forum that the existing rulebook for finance was built around a human making the final call, and that AI agents deployed at scale have already outrun that assumption. Sarah Breeden, the Bank's Deputy Governor for Financial Stability, used a panel at the ECB Forum on Central Banking 2026 in Portugal to argue that supervisory frameworks "were not built to contemplate autonomous agents." She framed those agents as software systems that take actions in markets, executing trades or moving payments, with little or no human sign-off.
The argument is narrower than it first sounds. It is not that AI is mysterious or that regulators have fallen behind a buzzword. It is architectural. The unit of accountability in finance is the human decision. When an autonomous system acts, the legal person who is supposed to be on the hook becomes harder to name, and the rulebook built around that person stops functioning the way it was designed to.
Breeden's specific point is that sign-off on every agent action is no longer realistic. As AI agents move from pilot projects into production trading and payments, a regulator or a bank's own compliance officer cannot review each transaction the way the existing model assumes. Reporting on the speech frames the consequence in starker terms. Agents reacting to the same market signal at the same instant can amplify shocks, in what supervisors call herding risk, and the cyber capability embedded in some agents raises the stakes further. Breeden told the forum that this combination could trigger "market meltdowns," according to coverage in the Business Times.
The signal that distinguishes this moment from earlier warnings is that supervisors have started naming the tools they are willing to use. Bank-to-bank recovery and resolution plans, the same playbook regulators apply when a lender fails, are now being discussed as a frame for AI-driven outages. So are emergency stop or "kill switch" mechanisms that would let a supervisor or a bank's own control room halt AI-driven trading when it threatens stability. These are options under discussion, not rules on the books, and the BoE has not proposed a specific legal form for either. A Memeburn summary of the policy direction treats the framing as gap-diagnosis rather than implementation.
Two parallel consultations show the gap is being treated as a durable policy problem, not a one-off remark. The Financial Stability Board, the international body that coordinates financial regulation across the G20, opened a June 2026 consultation on sound practices for the responsible adoption of AI in finance, with the full consultation report setting out what supervisors expect from banks deploying the technology. The UK government has also positioned itself as a home for AI-driven payments, with HM Treasury backing fintech adoption of future payment technologies in a parallel domestic track. That combination, international coordination plus a national adoption push, is what makes the BoE's gap-diagnosis consequential rather than rhetorical.
The deepest part of Breeden's argument is about legal form, not technology. Existing rules attach obligations to a person or firm: a bank, a board, a senior manager. When an AI agent acts, the action can be traced to a system, a model, and a deployment decision, but the person whose judgment authorised the action is several layers removed. Regulators know how to supervise a trading desk. They do not yet have a settled theory of how to supervise an autonomous agent that, in effect, is the desk. The kill-switch conversation is, in this reading, an attempt to retrofit accountability by giving supervisors an override on the system itself, rather than chasing the human at the keyboard.
There is one obvious caveat. Almost every authoritative quote in the public record traces to one official. The Breeden speech is the primary document, and the Business Times and Memeburn coverage amplify the same framing. No senior regulator at the SEC, ESMA, the FCA, or the Bank for International Settlements has, on the public record supplied here, used the same gap-diagnosis language. The picture is therefore a clean signal from one financial stability authority, not yet a chorus.
The next test will be whether the FSB consultation, when it closes, names autonomous-agent governance as a distinct workstream, and whether the UK Treasury consultation publishes a date for its own AI-in-finance track. Until then, the rulebook gap is named. It is not yet closed.