IMF's Adrian Warns AI-Driven Herding Could Trigger Flash Crashes in Global Markets

The Quick Take

• Outgoing IMF chief economist Tobias Adrian has warned that AI integration among market participants creates dangerous herding behavior that increases flash crash risk.
• Adrian's concern centers on the concentration of similar AI models among institutional investors amplifying synchronized trading responses to market signals.
• The IMF warning echoes earlier BIS research on AI-driven systemic risks in financial market microstructure and is likely to accelerate regulatory action.

Tobias Adrian's warning as he departs the IMF's research function represents a significant institutional signal about the systemic risk posed by AI's growing role in financial market microstructure. The herding concern is distinct from traditional algorithmic trading risks: it addresses the scenario where multiple independent AI systems, trained on similar data sets and optimization objectives, arrive at similar trading signals simultaneously — creating synchronized order flows that can overwhelm market liquidity and trigger flash crash events. Unlike human herd behavior, which operates at the speed of information, AI herding can aggregate positions across thousands of institutions within milliseconds.

The market implication is concentrated in high-frequency and institutional trading ecosystems where AI-driven execution is most prevalent, including US equity markets, European forex, and sovereign bond markets. For German financial institutions including Deutsche Bank, Commerzbank, and the major insurance groups, the IMF warning is relevant because they increasingly rely on AI-assisted risk management systems that share methodological similarities with peer institutions — precisely the concentration risk Adrian describes. Regulators at the ECB, BaFin, and the UK's FCA are likely to cite this IMF research in consultations on AI governance frameworks for systemically important financial institutions.

The forward regulatory signal is whether the Basel Committee on Banking Supervision or the Financial Stability Board incorporates AI herding risk into their macroprudential frameworks in the next annual review. Watch for ESMA's AI financial market surveillance guidance expected later in 2026, which will likely reference the IMF's herding concern as a justification for concentrated model usage disclosure requirements. The macro variable is volatility regime: AI herding risks materialize most acutely during low-liquidity periods and market stress events, making the VIX level at the time of any flash crash test the key observable for validating Adrian's thesis in real time.

Synthesized from 1 source.