The Bank of England is warning that AI trading agents could trigger the next market crisis

The Bank of England is warning that autonomous AI trading agents could trigger a market crisis by herding into identical decisions at machine speed, turning manageable stress into systemic collapse. Deputy Governor Sarah Breeden highlighted the risk of correlated AI behavior, and the Bank is running simulations to test how such behavior amplifies market stress. The Financial Policy Committee noted that firms' private incentives to deploy agentic AI may fail to internalize negative externalities, threatening overall financial stability.

Regulators at the Bank of England are sounding a specific alarm: AI agents executing trades autonomously could herd into identical decisions at machine speed, turning a manageable stress event into a systemic collapse. Sarah Breeden, the Bank of England's Deputy Governor for Financial Stability, has named the risk plainly. As AI agents take on more autonomous roles in executing trades and managing risk across financial markets, the danger isn't that they'll make bad individual decisions. It's that they'll make the same decision, at the same moment, faster than any human circuit breaker can respond. The Bank has confirmed it is running simulations specifically designed to test how correlated AI behaviour amplifies market stress, and the language coming out of Threadneedle Street is notably less hedged than what regulators usually allow themselves. Herding isn't a new phenomenon in finance. Quant funds have been blamed for flash crashes before, and risk parity strategies across the industry famously synchronized during the March 2020 COVID sell-off. But the concern with AI agents is structural, not incidental. When multiple systems are trained on similar data, optimizing similar objective functions, and responding to similar real-time signals, correlation isn't a side effect. It's baked in. The Bank's Financial Policy Committee put it directly in its April 2026 record: firms' private incentives to deploy agentic AI may fail to internalize the negative externalities, meaning what looks rational for each individual firm can still make the whole system less stable. The Bank isn't speculating about a distant future. It's already collaborating with international counterparts to model what cross-border herding behavior looks like, and the FCA launched its voluntary AI Live Testing service in April 2026 specifically to let firms trial models in controlled conditions before they go live at scale. That's a meaningful shift from principle-setting to active monitoring, even if formal rulemaking remains a step away. The FPC has asked for further work focused on agentic AI use cases in payments and financial markets, with a fresh joint Bank-FCA AI survey planned for later this year. Frankly, the timing matters. The BIS dropped its annual economic report on June 28, identifying an AI capital expenditure bust as one of three acute pressure points for global financial stability. The five largest hyperscalers are on track to spend more than a trillion dollars on AI infrastructure through 2026, and the BIS warned that circular financing structures mixing equity, debt, and supplier contracts carry assets that may be pledged multiple times over. If AI returns disappoint, those structures unwind, the same leveraged hedge funds now dominating sovereign bond markets face fire-sale pressure, and you've got a direct feedback loop from tech bust to sovereign debt crisis. The BOE's concern about agentic trading behavior and the BIS's concern about AI financing fragility are different risks, but they share the same trigger: a sudden, correlated reversal. The question for quant funds and algorithmic trading firms is whether they're already exposed in ways they haven't modeled. Most risk frameworks were built around human-speed decision cycles and assume some natural dispersion in how market participants respond to the same signal. AI agents break that assumption. If three major trading firms are each running agents that interpret rising volatility as a sell trigger, and those agents all act within the same millisecond window, the market impact isn't the sum of three firms selling. It's a coordinated event with no coordinator. There's a monitoring gap here that regulators have been honest about. The BOE acknowledged it doesn't yet have the tools to observe AI agent behavior in real time across the market. That's not unusual for emerging technology, but it means the first real test of this risk is likely to happen before the regulatory infrastructure to catch it is in place. The FCA's live testing regime helps at the firm level, but it doesn't give the Bank a system-wide view of how agents interact with each other across institutions. For fintech founders and capital allocators, Breeden's warning is an institutional credibility signal worth taking seriously. When a central bank deputy governor names a specific mechanism, not just a general category of AI risk but the specific dynamics of correlated autonomous agents amplifying stress, it's a sign that the supervisory conversation has moved past the exploratory phase. The BOE doesn't publish those concerns to calm markets. It publishes them to start shaping behavior before the event it's describing actually arrives. The harder part is that there's no clean solution on offer yet. You can't simply ban agentic AI from trading without handing a significant competitive disadvantage to regulated firms. You can mandate diversity in model architecture or training data, but that's technically complex to enforce and easy to circumvent. What you can do is build circuit breakers that don't assume human-speed markets, require firms to stress-test for correlated agent behavior specifically, and develop the real-time surveillance tools needed to spot herding as it forms rather than after the damage is done. The Bank knows this. Whether it moves fast enough is the open question. Also read: Call center giants are being repriced out of existence before AI has finished the job https://startupfortune.com/call-center-giants-are-being-repriced-out-of-existence-before-ai-has-finished-the-job/ • The AI notetaker sitting in your Zoom call may be your next legal liability https://startupfortune.com/the-ai-notetaker-sitting-in-your-zoom-call-may-be-your-next-legal-liability/ • SAP reorganizes its executive board around AI as investor patience runs thin https://startupfortune.com/sap-reorganizes-its-executive-board-around-ai-as-investor-patience-runs-thin/