European Central Bank examines systemic risks linked to AI-driven finance
Large language models showed less coordinated but more unpredictable behaviour, raising concerns about new forms of market uncertainty.
The European Central Bank (ECB) has published research examining how AI systems could affect financial stability as AI adoption expands across financial markets.
According to Eurosystem research, different AI architectures may produce significantly different market behaviours under similar economic conditions.
ECB simulations compared reinforcement learning systems with large language model-based agents operating in simulated financial environments. Researchers found that some reinforcement learning systems displayed coordinated responses resembling bank run dynamics in certain scenarios.
The report linked part of this behaviour to risk-avoidance patterns associated with prior negative outcomes.
Large language model-based systems showed lower coordination but more variable and unpredictable responses during periods of uncertainty. Despite receiving identical instructions, LLM-based agents frequently developed different assumptions about market behaviour, particularly during periods of moderate economic uncertainty.
ECB researchers noted that such inconsistency could create its own form of instability as AI-generated expectations diverge across financial markets.
The ECB suggested that wider AI adoption in finance may require updated risk management practices, investor awareness, and regulatory safeguards.
The research also highlighted the potential importance of existing market stabilisation measures, including circuit breakers and investor protection mechanisms.
Why does it matter?
AI is rapidly becoming embedded in trading, investment management, and financial decision-making across global markets, meaning flaws in AI behaviour could amplify systemic risks at unprecedented speed and scale.
The research signals that financial stability may increasingly depend not only on economic fundamentals and regulation, but also on the underlying architecture, coordination patterns, and predictability of the AI systems shaping market activity.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our chatbot!