The European Central Bank has published research on how Artificial Intelligence systems could influence financial stability as their use expands across financial markets. The work found that different Artificial Intelligence architectures can generate notably different market behaviours even under similar economic conditions, suggesting that the design of these systems may become an important factor in market dynamics.
Eurosystem simulations compared reinforcement learning systems with large language model-based agents in simulated financial environments. Some reinforcement learning systems showed coordinated responses that resembled bank run dynamics in certain scenarios. Researchers linked part of this behaviour to risk-avoidance patterns associated with prior negative outcomes. That finding indicates that Artificial Intelligence systems trained through repeated feedback could amplify stress if they react in similar ways during periods of market pressure.
Large language model-based systems appeared less coordinated but more variable and unpredictable during periods of uncertainty. Despite receiving identical instructions, these agents often formed different assumptions about market behaviour, especially during periods of moderate economic uncertainty. ECB researchers warned that this inconsistency could create a different kind of instability, as Artificial Intelligence-generated expectations diverge across financial markets rather than moving in lockstep.
The research points to a need for updated risk management practices, stronger investor awareness, and regulatory safeguards as Artificial Intelligence adoption widens in finance. It also highlights the continuing importance of market stabilisation tools such as circuit breakers and investor protection mechanisms. As Artificial Intelligence becomes more embedded in trading, investment management, and financial decision-making, financial stability may depend not only on economic fundamentals and regulation, but also on the architecture, coordination patterns, and predictability of the systems driving market activity.
