Bank of England Deputy Governor Sarah Breeden highlighted the need for bespoke AI regulation to manage risks to the financial system from increasingly capable agentic systems. Speaking at the European Central Bank Forum, Breeden noted that existing frameworks might be insufficient for autonomous agents, particularly as 52% of finance firms already use agentic AI, primarily for lower-risk tasks, a trend that could change rapidly. She emphasized that frameworks were not built for autonomous agents, and relying on a "human in the loop" for all agent actions is unrealistic.
Breeden expressed concern that if many AI agents, trained similarly, react identically to the same prompts or triggers, they could amplify volatility during stress, potentially leading to a market meltdown. Unlike human traders, who react at different speeds, synchronized AI agents could turn a minor fluctuation into a severe crash before human intervention is possible. This shared concern is echoed by the Financial Stability Board, which recently called for tighter safeguards against AI agents due to their distinct challenge to human oversight, particularly after Anthropic's Mythos model raised cybersecurity concerns for the banking industry.
To mitigate these risks, the Bank of England is considering measures such as fresh guardrails and "circuit breakers or kill switches" that could halt market-wide trading if faulty AI models cause a meltdown. Additionally, Breeden mentioned the exploration of "enhanced recovery" for core systems, allowing one bank to take over another's essential functions during disruptions. She stressed that international cooperation is crucial, as AI capabilities can spread globally, and warned against waiting for a crisis to build necessary cooperation.
