A new ECB study examines how AI agents impact financial stability, comparing Q-learning and large language models. Q-learning AI can trigger coordinated, unnecessary panics resembling bank runs, while LLMs lead to uncoordinated, unpredictable withdrawals due to differing interpretations. The article highlights that AI architecture fundamentally alters financial fragility, posing new risks despite AI's benefits. It recommends regulators understand AI tools, bolster risk management against AI speed, and expand financial literacy to cover AI delegation risks. The author emphasizes the need for proactive policies to prevent algorithm-triggered chaos, recalling the 1987 Black Monday crash.