Amid pressure to fight money laundering, Treasury looks to spur AI innovation

The Treasury Department will focus more effort on technical solutions to the country’s growing and ever-changing money laundering challenges, according to a strategy document the department published last week.

The strategy, last updated in 2020, primarily focuses on regulatory and operational aspects of the U.S.’s fight against money laundering and financing of terrorism. But Treasury added technological innovation as the fourth priority to the 2-year-old list. 

The Federal Reserve potentially developing its own digital currency, other virtual assets like stablecoins, and creating safe harbors or a regulatory “sandbox” for AI-driven products designed to fight money laundering and terrorism financing were specifically mentioned. Banks can use artificial intelligence to improve their anti-money-laundering processes and avoid issues with regulators. The technology also can help the government’s efforts in cracking down on money laundering.

“Data analytics and AI play an increasingly important role in informing policymakers of illicit finance threats and vulnerabilities by enabling agencies to sift through and synthesize vast quantities of data generated in financial crime investigations and analysis,” the strategy document says.

The strategy, which has a two-year timeline for implementation, comes as the Treasury faces scrutiny from lawmakers over its anti-money-laundering efforts amid heightened sanctions on Russia and an increase in suspected financial crimes driven in part by more crypto-related reports.

“Illicit finance is a major national security threat and nowhere is that more apparent than in Russia’s war against Ukraine, supported by decades of corruption by Russian elites,” Elizabeth Rosenberg, assistant secretary for terrorist financing and financial crimes, said in the press release on the strategy. “We need to close loopholes, work efficiently with international partners, and leverage new technologies to tackle the risks posed by corruption, an increase in domestic violent extremism, and the abuse of virtual assets.”

The agency is looking to incentivize testing and “responsible use” of AI for transaction monitoring, in part to ensure the efforts do not compromise the privacy and security of Americans’ data.

“Treasury is working to identify ways to combine financial technology solutions with policy and regulatory reform efforts to improve the overall effectiveness of private sector anti-money laundering and countering the financing of terrorism (AML/CFT) programs,” the strategy document says. “This includes the development of secure, consent-based, privacy-preserving digital identity solutions that include the necessary safeguards to minimize fraud and eliminate the incentives for fraudsters and hackers to steal personal data.”

Many banks partner with companies specializing in artificial intelligence to build the models and algorithms they need to fully capitalize on the gains the technology offers in fighting losses to financial crime. These companies include Feedzai, Nice Actimize, FiVerity, Inscribe, Resistant AI and larger firms such as IBM.

Among the specific benchmarks the Treasury aims to satisfy in the next two years are considerations for a safe harbor or regulatory sandbox “to nurture the development of new AI-driven AML/CFT products” — a response to concerns like those expressed last year by the Bank Policy Institute that regulators could stifle innovation of AI products with prescriptive regulations.

The existing regulatory framework is designed around older transaction monitoring systems that use a rules-based approach to flag suspicious activity, and some policy experts say that boxes banks into the rules-based approach that assures they do not run afoul of federal requirements.

Him Das, the acting director of the Treasury’s Financial Crimes Enforcement Network, said in January these sandboxes would allow financial institutions to transform that regulatory framework. Das also said Fincen is looking for banks to tell the agency: “What are you interested in building? What assurances do you need to start? What risks should we be on the lookout for?”

Treasury also said it would “consider issuing guidance to incentivize the testing and responsible use of AI for transaction monitoring,” but did not name any specific proposals.

One reason artificial intelligence offers financial institutions an edge is that it is particularly adept at identifying novel patterns. When fighting adversaries who learn, a system that is able to learn and adapt is also needed, according to Yaron Hazan, vice president of regulatory affairs at ThetaRay, an AI-driven transaction monitoring company.

“Criminal activity is unusual, but where is it unusual?” Hazan said. “We don't know, because the bad guys will search for the ways to find what they do to avoid being detected by the common methods.”

That is where artificial intelligence kicks in, he said. Unsupervised learning algorithms process transaction data and additional datasets that can provide context about those transactions. Those algorithms then cluster or categorize transactions, which can reveal hidden patterns or data groupings that fraud analysts might not be able to identify unaided.

“It can teach you about new use cases of money laundering and crime,” Hazan said. “That’s the beauty of it.”