Flash Crash

Definition

A flash crash is a market event characterised by a sudden, extreme price decline that occurs within minutes or seconds, typically followed by a rapid but often incomplete recovery. Flash crashes are caused by the interaction of multiple automated trading systems that respond to market signals faster than human participants can intervene. When one algorithm’s sell order triggers another algorithm’s automated response, which triggers further algorithms in a chain reaction, the result is a cascading collapse in prices driven entirely by machine-to-machine interaction. The phenomenon is a canonical example of multi-agent system failure: each individual algorithm behaves according to its specification, but their collective interaction produces a catastrophic emergent outcome.

How It Relates to AI Threats

Flash crashes are a documented threat within the Economic and Labor Threats and Agentic and Autonomous Threats domains. They demonstrate how autonomous AI agents operating in shared environments can produce emergent failures that no individual agent was designed to cause. As AI systems become more prevalent in financial markets — from high-frequency trading to portfolio management to credit decisions — the potential for AI-driven cascading failures increases. Flash crashes also illustrate the broader principle that multi-agent systems can exhibit unpredictable behaviour even when each agent individually functions as intended.

Why It Occurs

• Automated trading systems operate at speeds where human oversight and intervention are impossible in real time
• Multiple algorithms responding to the same market signals create correlated behaviour that amplifies price movements
• Stop-loss orders and risk management algorithms can create cascading sell pressure during a downturn
• The withdrawal of market-making algorithms during high volatility removes liquidity precisely when it is most needed
• Lack of coordination between independent algorithmic systems means no single participant can predict or prevent collective failure

Real-World Context

The most prominent flash crash occurred on May 6, 2010, when the Dow Jones Industrial Average lost approximately $1 trillion in market value in 36 minutes before partially recovering. The event was triggered by a large automated sell order interacting with high-frequency trading algorithms. Subsequent flash crashes have occurred in currency markets (2015 GBP flash crash, 2016 GBP and 2019 JPY flash crashes), bond markets, and individual equities. Regulatory responses include circuit breakers, minimum resting times for orders, and enhanced algorithmic trading oversight requirements (EU MiFID II). The proliferation of AI-driven trading systems raises concerns about increasingly complex and less predictable flash crash dynamics.