The Pentagon’s research arm is seeking new artificial intelligence tools that can uncover hidden flaws in the rules governing US financial and other digital markets and fix them before adversaries can exploit them to cause economic chaos.
The Defense Advanced Research Projects Agency’s new Intrinsic Market Resilience program is aimed at protecting electronic markets tied to everything from stocks and bonds to energy and cloud-computing resources.
Those critical systems have become faster and more efficient, DARPA says, but their increasingly interconnected nature has also created new avenues for disruption.
“I think that the incredible advances in automated reasoning and in AI are opening new ways to defend things,” Matthew Wilding, the IMR program manager at DARPA, told Business Insider. “They can be used to attack things as well, but the focus of this program is using them to defend things, to find those counterexamples and find those improvements we can make to these designs.”
“The program is taking these two technologies that have really advanced in the last few years and applying them to these electronic markets,” Wilding said, “so that they are more resilient and more dependable going forward.”
The concern is less that adversaries could hack these systems or steal data, both longstanding threats, than that they could exploit weaknesses in the design of the markets themselves.
The agency pointed to two historical examples of disruption.
In the early 2000s, traders exploited weaknesses in California’s newly deregulated energy market, contributing to a crisis that DARPA said cost $40 billion to $45 billion. That chaotic incident forced the federal government to invoke the Defense Production Act to ensure the supply of natural gas to military facilities, which DARPA said highlighted how the manipulation of market mechanisms could have serious consequences for military readiness.
Spencer Platt/Getty Images
The other example was the 2010 “flash crash,” which saw US stocks plunge within minutes amid market instability that investigators later linked in part to one individual’s manipulative trading practices, among other problems. Many of those gaps have since been corrected.
“The actions cost the individual nothing because the sell orders were specifically designed not to be executed by leveraging a quirk in the design of the market,” DARPA said in the solicitation posting for new AI tools.
The number of electronic markets and participants is on the rise, raising the risk of disruption and increasing the potential that exploitation could be deeply damaging. Adding to the problem is that the proliferation of AI and other new technologies has lowered the barrier to entry for bad actors.
IMR intends to leverage some of those same systems to model markets, analyze protections, and point out potential weak points.
The program isn’t aimed at the kinds of rogue market manipulation regulators typically pursue, such as individuals seeking financial profit. Instead, DARPA is focused on bad actors whose goal is simply to create disruption, economic damage, or instability, aided by current inadequacies in legal and regulatory safeguards.
The agency wants performers to use large language models to turn things like regulatory filings and technical specifications into computer models of how markets operate and interact.
The program aims to find ways to “identify design flaws in domestic online equity markets and recommend repairs” to protect against bad actors seeking to “exploit equity market design vulnerabilities,” DARPA said in the posting.
If a model can’t prove that a market meets a certain standard of resilience, DARPA wants it to generate a specific example of how the market could fail, allowing users to identify the weaknesses and propose a fix.
“We’re not just testing,” Wilding said, “We’re actually looking at it and analyzing it,” ultimately showing “how is it resilient, how it is not resilient, and how could we make it resilient.”
The DARPA solicitation reflects a growing national security concern that future attacks on the US might not resemble conventional military action. Future attacks could instead exploit vulnerabilities in digital infrastructure, supply chains, or markets.
