The U.S. Defense Advanced Research Projects Agency (DARPA) is exploring how AI can uncover hidden weaknesses in the rules governing electronic markets—and suggest fixes before hostile actors exploit them. The project spans markets for stocks, bonds, energy and cloud-computing resources. Its focus is not conventional hacking or data theft, but vulnerabilities in market mechanisms that could be used to cause disruption, economic losses or problems for military supply needs.
DARPA targets weaknesses in market design
The initiative, called Intrinsic Market Resilience (IMR), reflects DARPA’s concern that electronic markets have become faster and more efficient while growing increasingly interconnected. As the number of markets and participants rises, a flaw in one mechanism could have wider consequences. Advances in AI and other technologies may also lower the barriers to carrying out disruptive activity.
IMR program manager Matthew Wilding said progress in automated reasoning and AI offers new ways to strengthen defenses, but the same capabilities could also be used to attack. The project aims to apply these tools to market design: identifying counterexamples that could lead to a failure and proposing ways to address them, rather than relying only on general system testing.
California energy markets and the 2010 flash crash
DARPA cites two past episodes as examples. In the early 2000s, weaknesses in California’s newly deregulated energy market were exploited by traders. The resulting disruption was said to have caused losses of $40 billion to $45 billion. DARPA says the federal government had to invoke the Defense Production Act to ensure natural gas supplies reached military facilities. The episode illustrates how market manipulation can extend beyond commercial losses and affect military readiness.
The agency also points to the 2010 U.S. stock-market “flash crash,” when prices plunged within minutes. Investigators later concluded that the event was partly linked to the manipulative trading of an individual, alongside other problems. DARPA says certain market-design features allowed sell orders to have an effect without being executed, at little cost to the person placing them. Many of the related vulnerabilities have since been fixed.
Large language models to turn rules into market models
DARPA wants project teams to use large language models to convert materials such as regulatory documents and technical specifications into computer models showing how markets operate and interact. Among the goals is to identify design flaws in U.S. online stock markets and recommend changes that could reduce the risk of those weaknesses being exploited.
If a model cannot demonstrate that a market meets a specified resilience standard, the project aims for it to produce a concrete failure example. That could help analysts pinpoint the problem and assess possible fixes. Wilding said the work will examine not just whether markets are reliable, but where they are resilient, where they fall short and how they might be improved.
Focus on disruption, not profit-seeking violations
DARPA says IMR is not aimed at the profit-driven market manipulation typically pursued by regulators, such as misconduct intended to benefit an individual financially. Instead, the project focuses on actors seeking to disrupt markets, cause economic harm or create instability—and on gaps that may remain in existing legal and regulatory safeguards.
The initiative reflects growing U.S. national-security attention to risks in digital infrastructure. Future attacks may not take the form of conventional military action; they could exploit weaknesses in digital systems, supply chains or markets. For financial markets, DARPA’s proposed models are intended to expose potential vulnerabilities in market mechanisms. Which designs may need to change will depend on further analysis and validation.