The Pentagon's research arm wants artificial intelligence to find hidden weak spots in the rules that run big electronic markets, then fix them before an enemy uses them. The Defense Advanced Research Projects Agency, known as DARPA, calls the effort the Intrinsic Market Resilience program. It covers markets for stocks, bonds, energy and cloud-computing resources.
What DARPA is worried about
The concern is not mainly hacking or stolen data. Those are old threats. DARPA is more worried about flaws in how the markets are designed. These systems have become faster and more efficient, the agency says, but they are also more tightly connected, which opens new ways to cause disruption.
Two examples from the past
- In the early 2000s, traders exploited weaknesses in California's newly deregulated energy market. DARPA said the resulting crisis cost $40 billion to $45 billion.
- That crisis forced the federal government to use the Defense Production Act to secure natural gas for military facilities.
- In 2010, the "flash crash" sent US stocks plunging within minutes. Investigators linked it in part to one person's manipulative trading, among other problems.
DARPA said many of those gaps have since been corrected. In the solicitation, it noted that the manipulative sell orders cost the trader nothing, because they were designed not to be executed. That worked because of a quirk in how the market was built.
How the AI would work
DARPA wants teams to use large language models, the same kind of AI behind popular chatbots, to read regulatory filings and technical specifications. The AI would turn them into computer models of how markets operate and interact. Those models would then be checked for weak points.
If a model cannot prove a market meets a certain standard of resilience, it must produce a specific example of how the market could fail. People could then see the weakness and propose a fix. Program manager Matthew Wilding said the goal is showing how a market is resilient, how it is not, and how to make it so.
Who the program targets
The program is not aimed at the usual rogue traders who chase profit. DARPA is focused on bad actors who simply want disruption, economic damage or instability. The agency says AI and other new technology have lowered the barrier for such actors, while the number of electronic markets and participants keeps growing.
What is not known yet
The material does not say which teams will take part, how much the program will cost or when results might arrive. It also does not describe any specific flaw that has been found. The solicitation does reflect a wider national security worry that future attacks may target digital infrastructure, supply chains or markets instead of looking like conventional military action.
The takeawayMarkets run on rules, and DARPA wants AI to find the flaws in those rules before an adversary does.