Bank of England Warns AI Agents Now Operating Financial Markets Unsupervised

Published by James Harris on July 1, 2026July 1, 2026

The Bank of England’s deputy governor for financial stability, Sarah Breeden, used her platform at the European Central Bank’s annual forum in Sintra on June 30 to issue a direct warning: autonomous AI systems are advancing faster than regulators anticipated, and financial infrastructure, including trading, payments, and cybersecurity, is now inside the blast radius.

The acceleration Breeden described is not abstract. In 2019, the length of software tasks that leading AI models could complete was doubling every seven months. By 2024, that doubling was happening every four months, and breakthroughs in cyber vulnerability detection this spring suggest the pace has quickened further. The trajectory she outlined runs from generative AI that responds to prompts, through reasoning models trained on multi-step problems, to agentic systems that now plan and execute decisions with zero human oversight. For financial markets, that means AI agents capable of trading securities, processing payments, and responding to threats before any human has seen the alert. The concern is not hypothetical novelty but a structural shift in who, or what, is actually operating the system. Efforts to create governance frameworks for these agents, such as open standards for documenting AI agent behavior in dispute contexts, are emerging precisely because the gap between capability and accountability is widening fast.

Breeden singled out cybersecurity as the most critical near-term risk, noting that the same agentic tools that help defenders find and patch vulnerabilities also give attackers the ability to discover and exploit them.

Her remarks carried two implications that go beyond the standard regulatory call-to-arms. First, she floated the idea that autonomous trading tools may need built-in kill switches to prevent market shocks, a meaningful departure from current regulatory frameworks that would require significant coordination across jurisdictions. Second, she flagged a financing risk that is less discussed: the Bank’s Financial Policy Committee concluded in April that large technology companies are increasingly funding AI infrastructure through debt rather than equity and cash flow, and that debt is taking on complex new forms. A sharp drop in AI-related asset valuations could now transmit stress through credit markets in ways that were not possible when the same infrastructure was equity-funded. The committee said the financial stability consequences of any fall in AI-related asset prices could well increase.

A more detailed assessment from the Financial Policy Committee on this topic is expected on July 7, which will likely give markets and regulators their first formal look at how the Bank of England intends to quantify and frame that contagion risk.