The Bank of England's deputy governor has warned that autonomous AI trading agents could trigger a market meltdown, calling for a "kill switch" mechanism to halt trading in extreme scenarios. Sarah Breeden told a central banking conference in Portugal last week that agentic trading "poses a threat to financial stability," as startups push AI agents into live trading without human supervision.
Two firms illustrate how quickly the practice is gaining ground. EquiLibre Technologies, founded by former Google DeepMind researchers in Czechia and now valued at $500 million, uses methods developed for poker algorithms to trade stocks. The company told TechCrunch its agents have delivered a profit in every single month since launch. New York-based Moment, created by ex-Citadel staff, is building infrastructure for wealth managers to deploy AI trading agents in fixed income and equity markets.
Until recently, AI agents in finance handled operational tasks - research, data analysis - and rarely operated autonomously. Regulators are now bracing for a shift as more firms move toward fully unsupervised execution.
The herding risk and the case for a kill switch
Breeden's central concern is that large numbers of AI trading agents will respond in near-identical ways to the same events. "Such 'herding behaviour' could amplify volatility in the market at times of stress," she said. A sudden, coordinated sell-off driven by faulty models could outpace human intervention, making some form of mandatory circuit breaker essential.
"Some sort of 'kill switch' could be required to 'limit or stop trading … if faulty AI models cause market meltdown,'" Breeden said. The Bank of England is working with the German Bundesbank and the Bank for International Settlements to shape a policy response.
Cyber threats compound financial stability risks
Breeden also highlighted the danger from advanced AI cyber capabilities, pointing to Anthropic's recent release of its "Mythos" model, which the company deemed too dangerous for general use. "In malicious hands, [these models] materially increase the chance of attacks that could harm financial stability," she said.
She stressed the need for rapid patching of cyber vulnerabilities and more frequent scenario planning to prepare for AI-driven attacks targeting critical financial infrastructure.
Why this matters for finance professionals
Regulatory expectations around algorithmic trading controls are set to rise. Risk managers and compliance officers will need to assess whether existing circuit breakers and model governance frameworks can handle autonomous agents. For those building or overseeing trading systems, a deeper technical grasp of AI behaviour - including how models can converge on correlated strategies - is becoming essential. Relevant AI for Finance Courses can provide that grounding.
Policy and governance teams face a parallel challenge. Understanding the regulatory direction and contributing to internal controls around agentic systems will require fluency in AI risk analysis. Specialised programs, such as AI for Policy Makers Courses, are designed to bridge that gap. The Bank of England's warnings signal that the window for preparation is narrowing.
Your membership also unlocks:
