Bank of England warns autonomous AI agents could amplify market stress

Bank of England Deputy Governor Sarah Breeden said the use of autonomous AI agents "could amplify volatility in periods of stress" in financial markets, and suggested that tougher regulation may be needed to address it. She made the remarks at the European Central Bank's annual symposium in Sintra, Portugal.

What exactly worries the regulator

The concern is not trading algorithms in general, which have been part of markets for decades. It is specifically autonomous AI agents: systems that decide on their own to buy, sell, or reallocate assets with minimal human involvement. Breeden points to the risk that many such agents, trained on similar data and following similar logic, could start acting in sync during market stress, amplifying price swings rather than dampening them.

Financial regulators have known about this "herding" problem since the early days of algorithmic trading. When dozens of market participants run similar models and react the same way to the same news trigger, a local sell-off can turn into a cascading crash. AI agents raise the stakes: their decision-making is far less transparent than the rules behind classic algorithmic trading, and they react orders of magnitude faster than humans.

"The use of autonomous AI agents could amplify volatility in periods of stress,"

Sarah Breeden said.

Why regulators are raising this now

Central banks and supervisors have spent years tracking the role algorithmic trading plays in sudden crashes and volatility spikes, the so-called "flash crashes." More autonomous AI agents, which can interpret news and reprice positions on their own without a trader, add a new layer of unpredictability. These systems react to market events many times faster than humans, and their decision-making is often opaque even to the people who built them.

• The statement came from Sarah Breeden, Deputy Governor of the Bank of England
• The venue was the European Central Bank's annual symposium in Sintra, Portugal
• The risk, in her words: AI agents' ability to "amplify volatility in periods of stress"
• Breeden allows that tougher regulation of autonomous AI agents in finance may be necessary

How regulators might respond

Central banks already have experience regulating algorithmic and high-frequency trading, using automatic circuit breakers, order-volume limits, and requirements to test algorithms before they go live. It is reasonable to expect similar tools to serve as the starting point for the debate on regulating AI in finance: disclosure of decision-making logic, mandatory model stress tests, and caps on the share of assets managed by autonomous agents. The difficulty is that modern AI agents, unlike classic trading algorithms with clear-cut rules, make decisions based on models whose logic is not always explainable even to their developers, which complicates the traditional supervisory approach.

The Sintra symposium is one of Europe's key forums where central bankers exchange assessments of risks to the financial system each year. The fact that AI agents came up there suggests regulators see the issue not as hypothetical but as pressing.

What it means

The remarks from a senior Bank of England official show that autonomous AI agents have moved from tech news onto the agenda of financial regulators. The next step could be concrete requirements for banks and funds to oversee systems that make decisions without human involvement.