Agentic AI may require regulatory reform, BOE’s Breeden says

A Central Banker's Warning on Autonomous AI

Sarah Breeden, a deputy governor at the Bank of England, has raised a pointed concern about the next wave of artificial intelligence entering financial markets: agentic AI systems that can act independently, make decisions, and execute transactions with minimal human oversight. Speaking at an event hosted by the European Central Bank, Breeden suggested that existing regulatory frameworks may not be equipped to monitor or contain the risks these systems pose to financial stability.

Why Agentic AI Is Different

Most AI tools currently deployed in finance operate as decision-support systems — they analyze data, flag anomalies, or generate recommendations that a human ultimately approves. Agentic AI represents a meaningful shift: these systems are designed to pursue goals autonomously, chaining together tasks like a trader might, without requiring a human to sign off on every action. In theory, this improves efficiency and speed. In practice, it introduces a supervisory blind spot. If an autonomous system makes a flawed judgment call during a period of market stress, the traditional chain of accountability — who approved the trade, who monitored the risk — becomes murkier.

Why This Matters Beyond the UK

Breeden's comments land at a moment when financial regulators globally are grappling with how to classify and supervise AI-driven decision-making. Her remarks echo concerns already embedded in the EU AI Act, which categorizes certain high-risk AI applications — including some financial services use cases — for stricter scrutiny, transparency requirements, and human oversight obligations. As agentic systems proliferate, enforcement bodies tasked with implementing the Act will likely face pressure to clarify how autonomous decision-making chains fit within existing risk tiers.

In the United States, where AI regulation remains more fragmented and sector-specific, Breeden's warning adds to a growing chorus arguing that financial regulators — the Federal Reserve, SEC, and OCC among them — may need new supervisory tools rather than relying solely on existing model-risk-management guidance designed for simpler algorithmic systems.

The Broader Policy Challenge

This intervention fits a pattern seen across AI safety policy circles: regulators increasingly acknowledge that rules built for static, human-supervised software don't map cleanly onto systems capable of independent, multi-step action. Central banks in particular worry about systemic risk — the possibility that many institutions deploying similar agentic tools could inadvertently create correlated behavior, amplifying shocks during volatile periods.

What Comes Next

Breeden did not detail specific reforms, but her remarks signal that financial supervisors are beginning to treat agentic AI as a distinct regulatory category rather than an extension of existing algorithmic trading oversight. Expect this theme to feature more prominently as the Bank of England, ECB, and other regulators refine supervisory expectations over the coming year.