BIS warns of AI herding: what happens when every firm uses the same model

Global reference

• Body: Bank for International Settlements (BIS)
• Speaker: Tao Zhang, Chief Representative for Asia and the Pacific
• Speech: Artificial intelligence, monetary policy and financial stability, Financial Street Forum 2026, Beijing
• Key date: Published by the BIS on 21 October 2026
• Status: Views are the speaker’s own. No specific measure is required

The risks he named

• Faster price changes: AI may let firms reprice more often, changing how inflation behaves.
• Herding: similar models, data or decision rules can lead to “correlated responses” that “could reinforce one another”.
• Leverage: growing debt, including private credit, and circular financing within the AI sector.
• Cyber and concentration: shared technology and critical providers mean one disruption could hit many institutions at once.

What he asked authorities to do

• Look at shared exposures, including reliance on common models and providers, not only single firms.
• Run scenarios: a sharp repricing of AI-related assets, a disruption at a key provider, or similar model-driven decisions under stress.
• Strengthen oversight of critical third-party providers, and keep human judgment central.

What to do now

• List which AI models and providers sit behind your forecasting, credit, pricing and fraud tools.
• Run one scenario where your main AI provider is unavailable for a week.
• Ask whether key models would push you to act like everyone else in a sell-off, and who can override them.
• Keep a manual fallback for cash forecasting and payments.

Our take

Central bankers have spent a month on AI debt. This speech adds a quieter risk: sameness. Using a popular model is not wrong, but finance teams should know when they rely on one, and keep the ability to decide differently.

Important. This summary is for general information and is not investment, regulatory or financial advice. It is based on public Bank for International Settlements publications and may be incomplete or out of date. The official publications prevail; check the official source before acting.