DARPA asking for ideas on automating money laundering detection
With all the AI hype swirling around, you'd think someone would've cracked this one already
Tracking down and preventing money laundering is a slow, time-consuming, manual procedure. DARPA is hoping it can provide some relief for exhausted analysts by automating the process.
The Pentagon's Defense Advanced Research Projects Agency announced the Anticipatory and Adaptive Anti-Money Laundering (A3ML) program last week with news it's seeking research proposals for "rapid graph-search algorithms" that are able to sift through financial transactions for suspicious patterns.
A3ML doesn't just aim to sniff out shady dealings, though: DARPA is also calling for algorithms that can anticipate future activities, aiming to transform anti-money laundering (AML) from a reactive to a proactive process – all while reducing privacy risks by avoiding the sharing of sensitive financial data.
"The way we address these threats today is largely through manual data collection and analysis methods that threaten privacy," said A3ML program manager David Rushing Dewhurst. "We are looking for new technological ideas that preserve privacy while setting a course to end our adversaries' financial warfare."
The need for such technology is a serious one. North Korea, infamous for its financial crimes, stole $659 million in cryptocurrencies in 2024, much of which has likely been laundered through various methods. That doesn't include fiat money laundering conducted on behalf of North Korea. Thankfully, Dewhurst told us that both fiat and crypto currencies are within A3ML's scope.
The US government estimates that half of North Korea's nuclear program is funded by laundered cash, DARPA said. Beyond the Kim regime's penchant for financial crime, a federal indictment alleges that money launderers tied to Chinese underground banking provide financial services to Mexico's Sinaloa cartel, a concern noted in the context of the A3ML program.
"Money laundering finances our adversaries' weapons programs, global terrorism, and the illicit drug trade, all of which threaten U.S. national security," Dewhurst said.
The Basel Institute on Governance, which has tracked progress in fighting money laundering since 2012, said in its 2024 year-end report that, while compliance is up around the globe, the effectiveness of AML measures has actually declined in recent years. According to the Institute, AML measures are only effective 28 percent of the time, down from 30 percent in 2021.
In short, something's gotta give, and it might as well include some new technology to improve effectiveness – and put a layer between analysts and raw, private financial data.
Per DARPA, the proposal specifically aims to develop algorithms that "represent patterns of illicit financial behavior in a concise, machine-readable format that's also easily understood by human analysts." Results would be extracted from financial databases "without directly sharing sensitive financial data," though the particulars will be up to participants in the program to figure out.
"A3ML technology will be designed to share illicit financial behavior templates instead of sharing sensitive financial information," Dewhurst told The Register. In other words, the intent is to establish a list of tactics, techniques, and procedures used by money launderers, not share specific info.
- Laundering cash from healthcare, romance scams lands US man in prison for a decade
- Dark web crypto laundering kingpin sentenced to 12.5 years in prison
- Anti-money laundering bill targeting cryptocurrency introduced in US Senate
- Feds reach for sliver of crypto-cash nicked by North Korea's notorious Lazarus Group
As to whether full-fledged AI models are being considered, Dewhurst told us "any technical solution that can enable the algorithmic extraction of financial tactics, techniques, and procedures from heterogeneous transaction data" is on the table.
"If widely adopted, A3ML would lower compliance costs and risks for industry, increase the accuracy and precision of illicit finance data available to the US government, and drastically reduce the sharing of Americans' sensitive financial information," Dewhurst added in an email to The Register.
As to when A3ML solutions could actually start stopping real-world money laundering – this is a research project, after all – don't expect immediate results. A3ML aims to fill the AML operational gap "as quickly as possible," Dewhurst told us, but Proposers Day for the project isn't happening until February 20, with actual proposals not due until March 10.
After that, it remains unclear what researchers actually come up with, so no rest for the weary AML analysts. There are countless rows of records to pour through still. ®