Agentic AI accelerates cyber risks, challenges financial stability

Bank of England Executive Director Sarah Breeden warned that accelerating agentic AI capabilities pose a double challenge for central banks and financial stability. Speaking at the ECB Forum, Breeden highlighted cyber risks, agentic trading, and payments as key areas of concern.

The accelerating pace of autonomous AI

AI capabilities are increasing exponentially, with the speed of software task completion doubling every four months in 2024, and potentially faster now.

This acceleration has led to an inflection point where generative AI, initially creating content, evolved to reason through requests, and now, with agentic AI, systems can autonomously chain together sequences of actions.

This means the financial system is likely to operate with greater autonomy, scale, and speed, with AI agents executing trading strategies, identifying cyber vulnerabilities, and transacting on behalf of consumers.

Breeden noted that the Bank's Financial Policy Committee (FPC) will publish an updated assessment on 7 July, which will likely highlight rising debt financing for AI infrastructure and increasing financial stability consequences from any fall in AI-related asset prices.

Cyber resilience and agentic trading concerns

Breeden's most immediate concern is the "step change" in agentic AI's cyber capabilities, which, in malicious hands, could materially increase attacks harming financial stability.

The challenge is to maintain the advantage of cyber defenders, especially given the rapid evolution of open-source models and the urgency of patching vulnerabilities across the financial sector and critical infrastructure.

This necessitates a system-wide approach to operational resilience, with authorities placing higher likelihood on simultaneous multi-firm disruption and enhancing tools for stress testing.

Furthermore, agentic trading in financial markets could amplify volatility if AI agents respond similarly to prompts, potentially leading to herding behavior.

The Bank of England is experimenting with simulation methods to understand these risks and explore mitigants like circuit breakers or kill switches.

Regulation for an autonomous future

The rise of agentic payments and commerce, where AI agents automate transactions like booking holidays or refilling fridges, also presents significant challenges.

Central banks must ensure new retail payment infrastructure supports these journeys, but the biggest issues lie in regulation and industry standards.

Questions arise regarding secure consent, liability for erroneous transactions, and avoiding fragmentation as AI firms develop their own protocols.

Breeden concluded that existing, technology-agnostic regulatory frameworks may no longer suffice for autonomous agents, necessitating more sophisticated governance and accountability frameworks.

Cooperation with other official sectors and internationally, including the FSB and G7, is vital to prepare for these frequent technology surprises.