Bundesbank Warns: Thousands of AI Treasury Agents Acting in Sync Could Trigger Systemic Risk

Bundesbank Flags Herding Risk Among AI Treasury Agents

The Bundesbank warned in its September monthly report, released on September 21, that if thousands of AI treasury agents make decisions simultaneously based on similar models, data, and objectives, they could trigger herding behavior and strain financial infrastructure.

The report noted that when these agents simultaneously choose to release payments, hedge foreign exchange risk, or cut counterparty exposure, individually optimal decisions could escalate at the systemic level into payment congestion, sudden liquidity tightening, and infrastructure stress. Agents delaying payments to preserve cash, or accelerating trades in response to the same signal, could both trigger such chain reactions.

The warning comes as AI treasury agents increasingly take over corporate treasury functions, gaining the ability to autonomously decide payment timing, where to hold funds, liquidity levels, and payment channel selection, ultimately responding to changes in exchange rates, interest rates, or counterparty risk. Corporates can issue instructions to agent systems—such as maximizing yield, maintaining liquidity, or minimizing transaction costs—and the agents can execute without human approval.

The Bundesbank recommended that CFOs establish limits, escalation triggers, and human checkpoints—not only to govern what agents are permitted to execute, but also to define at what scale otherwise reasonable automated decisions become dangerous.

Source: https://www.pymnts.com/news/artificial-intelligence/2026/what-happens-when-10000-treasury-agents-make-same-decision/