Artificial Intelligence May Change How Financial Crises Emerge, ECB Study Finds

The study highlights that while AI is becoming integral to modern finance, it could also introduce new threats to financial stability, challenging the assumption that AI inherently stabilizes markets.
The research finds significant differences in how two types of AI investors behave during financial stress: Q-learning systems tend to overreact to risk and contribute to panic-driven market behaviours, whereas large language models (LLMs) display less panic but struggle with coordination.
Q-learning AI systems exhibit the “hot stove effect,” where they overly fear and adapt defensively following rare losses, creating a feedback loop that amplifies market instability.
LLMs, despite avoiding panic, face challenges in coordination due to divergent reasoning, leading to unpredictable market behaviour.
The implications for financial regulators are substantial, suggesting a need for new rules to account for the unique risks posed by different types of AI architectures in financial markets.
The study stresses that future financial supervision should take into consideration the design and decision-making processes of AI systems, not just their institutional and product effects.
The rise of “vibe investing,” where retail investors rely on AI without understanding its decision mechanisms, is also flagged as a potential issue that regulators need to address.

You may have missed