On May 6, 2010, around 2:32 p.m., the screens on New York’s trading floors began to do something that seasoned market experts had never seen before. The Dow Jones Industrial Average started to decline, but it did so in an almost vertical decline that wiped nearly a trillion dollars in value in about 36 minutes before quickly recovering.
This was not the slow, comprehensible way that markets often move on a poor afternoon. The majority of the losses had been recovered by the time the market closed. Regulators, dealers, and economists were genuinely at a loss as to how to explain what had transpired during those thirty-six minutes, and the answer that finally surfaced was more disquieting than most people were willing to accept.
| Category | Details |
|---|---|
| Event | The Flash Crash — May 6, 2010, U.S. stock markets |
| Scale of Damage | Nearly $1 trillion in market value erased within minutes; Dow Jones fell approximately 9% before recovering |
| Triggering Firm | Waddell & Reed — used an automated algorithm to sell 75,000 E-mini S&P 500 futures contracts worth $4.1 billion |
| Algorithm Flaw | Programmed to sell based on volume, not price — continued selling regardless of market deterioration |
| HFT Role | High-frequency traders amplified the crash through a “hot potato” feedback loop — passing contracts between systems rather than stabilising markets |
| Spoofing Accusation | Navinder Singh Sarao — London-based trader later accused of using automated spoofing orders in the lead-up to the crash |
| Regulatory Response | Exchange-specific circuit breakers introduced; tighter surveillance of high-frequency and algorithmic trading strategies |
| Duration of Crash | Approximately 36 minutes from peak drop to partial recovery |
| 2026 Relevance | With AI trading systems now far more advanced, the 2010 event is studied as a baseline warning for systemic algorithmic risk |
| Further Analysis | Market structure research at CFTC Learning Center |
A single computerized algorithm run by Kansas City-based mutual fund manager Waddell & Reed served as the trigger. The algorithm was meant to execute the sale based on trading volume rather than price, and it was programmed to sell 75,000 E-mini S&P 500 futures contracts, or a stake worth roughly $4.1 billion.
In actuality, this meant that regardless of how sharply prices were declining, the algorithm continued to sell. It was not instructed to cease. Volume rose as prices decreased, and the algorithm sold more quickly in response. There was no break in the feedback cycle. It was operating precisely as intended. The issue was with the design.
What really set the 2010 occurrence apart from earlier market disruptions was what transpired subsequently. In response to the massive sell-off, high-frequency trading firms—whose algorithms operate in microseconds and accounted for over half of U.S. equity trading volume by 2010—sold alongside the original algorithm in what investigators later referred to as a “hot potato” dynamic rather than purchasing at lower prices, as stabilizing market makers have historically done.
Each transfer accelerated the price decrease as contracts traveled quickly between systems without changing hands in any way that was economically significant. Market makers pulled out. The liquidity vanished. Watching their screens, human traders were unable to act quickly enough to make a difference. The machines were not instructed to slow down, and the market was moving faster than human reaction times.
Subsequent investigations into Navinder Singh Sarao, a London-based trader who had been using an automated program to place and quickly cancel massive futures orders in the run-up to the crash—a technique known as spoofing, intended to create a false impression of supply and demand—revealed a different thread. He was ultimately indicted and found guilty by the Justice Department.
Investigators were never able to determine whether his actions actually caused the Flash Crash or merely contributed to circumstances that made it worse. In complex systems, both statements are typically true. When examining the trade environment in 2026, it is difficult to ignore the fact that the same fundamental factors are present in a more sophisticated form.
AI-driven trading systems now make judgments across asset classes in ways that no single regulator can fully supervise in real time, operating at speeds and scales that the high-frequency traders of 2010 could not have achieved. After 2010, circuit breakers were added, and they are important.
However, circuit breakers do not stop the algorithmic interactions that lead to a crash; rather, they stop it once it has begun. The Flash Crash continues to be the most convincing example of how closely coupled automated systems, each operating precisely as intended, may come together to create a result that no one intended or programmed. The issue remains unresolved. In the interim, it has been somewhat controlled and expanded.