Guest Post: “Is Artificial Intelligence the Key to Fixing America’s Money Laundering Problem?”

Today’s post is by Duke Law’s Mr. Robert DeNault who explains in how artificial intelligence (AI) offers a tool to help tackle the extremely difficult problem of stopping the scourge of 21st century financial systems: money laundering.

As some context, consider what the U.S. State Department observes

“Money laundering is the process of making the proceeds of criminal activity appear to have been legally obtained. According to the IMF and World Bank, criminals launder an estimated two to nearly four trillion dollars each year.

“Among those who seek to disguise the illegal proceeds of their crimes are drug traffickers, terrorists, corrupt public officials, and organized criminal groups. Introducing illegally obtained funds into the stream of legitimate commerce and finance allows criminals to profit from their illegal activity, taints the international financial system, and erodes public trust in the integrity of the system.”

“Some criminals use the financial system to support terrorists or acts of terrorism. Terrorist financiers and other criminals use the formal financial system, new payment methods such as bitcoin and Ripple, traditional methods of value transfer such as hawala*, trade based money-laundering, and cash couriers, particularly in countries with non-existent or weak national anti-money laundering/countering the financing of terrorism (AML/CFT) tools.”  (Emphasis added.)

Mr. DeNault’s name will be familiar to many Lawfire readers.  You may recall his brilliant paper, “The Crisis of Cryptocurrency: Executive Branch Authority to Address the World’s Most Potent Financial Threat“, was the inaugural offering of the our online “Essays on Law, Ethics and National Security – A LENS Center Series.”  Obviously, he is someone to watch in the years ahead as he is establishing himself as one of the most thoughtful thinkers as to how financial systems might be strengthened to prevent exploitation not just by sophisticated criminals, but also by terrorists and others who threaten national security.

Mr. DeNault fully grasps a point I’ve longed believed is oft underappreciated: the nexus between national security and commercial activities (hence our recent series on opportunities for those interested in national security, but who are also fascinated by private sector business and financial matters.)  I’m also extremely pleased to tell you he’s one of my much-valued research assistants!

What I especially like about his discussion is that it cogently and concisely gives you a useful overview of a subject that could easily become mired in complexity and, ultimately, confusing detail.  (His future clients will surely appreciate this hard-to-find talent!)  Anyway, here’s his essay:

Is Artificial Intelligence the Key to Fixing America’s Money Laundering Problem?

By Robert J. DeNault

The movement of illicit money through financial systems enables bad actors to continue unlawful activity while it corrupts markets wholly dependent on their perceived integrity. In recent years, the publication of caches of secret documents by international journalists—like the Panama Papers and the FinCEN Files—highlighted the vast scope of the world’s money laundering problem: millions of shell companies, owned by other shell companies, making loans to one another and doing business with global banks in order to convert ill-gotten assets into real capital hidden in offshore accounts. Money laundering is not just a rampant crime; it has turned into its own industry.

But why are journalists the ones exposing this worldwide financial pandemic? Where are the regulators, the lawyers, the prosecutors? One reason that journalists took the lead on exposing this problem is that the system for addressing money laundering is outdated and overwhelmed. Banks have flagged millions of suspicious transactions, giving regulators and investigators a mountain of information to sort through in order to determine which are legitimate cases of money laundering and which are false flags. A lack of specific money laundering regulations for complex financial institutions or new forms of digital currency is further hindering the law enforcement effort to follow the money. But novel developments in artificial intelligence (AI) present endless possibilities to resolving some of these critical issues.

Over the past two decades, the U.S. federal government vastly expanded the toolkit for government and private sector actors to stop money laundering. New reporting requirements for banks, expanded authorities for law enforcement agencies, and harsher penalties for offenders represented a recognition that money laundering was becoming a critical problem. And just last week, Congress passed a new law requiring the disclosure of beneficial owners of secretive shell companies in an effort to improve detection of money laundering.

But it has not been enough. The U.S. financial system is still riddled with illicit money: hedge funds, internationally important banks, commercial real estate giants and even health care systems have become laundromats for tainted foreign money. If law enforcement cannot get further in front of efforts to wash dirty money through our financial institutions, it is a matter of time until the integrity of the entire financial system comes into question. Novel AI systems offer methods to improve detection and prevention systems at the first lines of defense: banks and government financial agencies.

Policing money laundering is a unique process. At the risk of oversimplifying, the U.S. money laundering detection and prevention system is largely split between banks and the government. Compliance officers inside the banks act like tipsters, flagging transactions and bad actors for regulators. They send their “tips” to Treasury regulators in the form of Suspicious Activity Reports (SARs).

SARs look like raw intelligence memos. They have a few sections for compliance officers to check certain categories of transaction types, and then longer sections where officers might detail their rationale for flagging a particular transaction. They may note that a particular withdrawal or new line of credit is engaging in odd activity; they might describe the history of the account holder and possible connections to criminals or terrorists.

This creates a unique structure of reporting already partially informed by AI. Current systems use AI techniques to search a large pool of bank transactions and account histories, identify client customers and analyze data that sits in segregated systems at a bank.

But technological experts from Ireland to Spain to Australia to the U.S. have published extensive work indicating that existing artificial intelligence systems are falling short in money laundering compliance. Some of their chief critiques: current systems are too rudimentary or not being applied in the most effective way; the same systems are missing suspicious transactions at Investment Banks and other complex financial institutions; and almost all operating AI systems are unable to address novel forms of transactions like cryptocurrency.

In fact, several computer science academics have published pieces specifically proposing AI solutions to three of the problems mentioned above: reducing the backlog of SARs, implementing more sophisticated AI software to detect money laundering at Investment Banks, and crafting better AI systems to detect fraud and laundering on cryptocurrency blockchains.

• • Reducing the backlog: Two researchers at the University of Lisbon in Spain propose new data mining approaches for profiling bank clients in order to improve detection of money laundering operations. Other researchers in Australia posit that developing entirely new AI systems focused on analyzing group behavior using a combination of network analytics and supervised learning techniques would help reduce the amount of false positives in suspect transactions and shrink the turnaround time on investigating a transaction from 1 week to a matter of minutes. Finally, private sector analysts have explored a new paradigm of data labeling and data collection to offer a much-needed update to the systems banks use to identify suspicious clients. Each of these measures would reduce the time compliance officers spend determining whether a flagged transaction is truly an instance of financial criminal activity, and thus reduce the backlog of SARs which grows larger each year.

• • Targeting Investment Banks: Scholars at the University College Dublin School of Computer Science adopted a similar approach and proposed a unique calibration of techniques to apply to investment banking. They sought to extract and clean raw datasets from data sources located in different sites of an international bank, and subsequently integrate them into consolidated databases used to build a data warehouse of customer information and customer transactions. These data mining practices not only identify suspicious activity from a broader scale and narrow it down—more efficient than the reverse—but also eliminate weekslong investigations performed almost entirely by beleaguered compliance officers. The researchers found that this approach was uniquely suited to flagging suspicious transactions at Investment Banks.

• • Tackling Cryptocurrency: Research done in the cryptocurrency laundering space advocates for better visualization of data detecting stolen bitcoins in the blockchain, using classification rules to detect laundering, and better detection of mixing services through data mining. Newer AI techniques like deepwalk and node2vec—unsupervised learning techniques that involve analyzing a network to make inferences based on neighborhood connections—significantly outperformed manually created tools used to detect cryptocurrency laundering. Other innovative temporal-based networks designed to investigate patterns of interaction between addresses might also be a key to unwinding dangerous mixing strategies employed by firms like Dark Wallet, which combine suspect and non-suspect transactions on the blockchain to confuse those who track cryptocurrency transactions. Investigators untangling webs of cryptocurrency laundering have spent years tracing bitcoin transactions to their original source, but these advancements may make their analyses significantly easier to undertake.

If these kinds of solutions are implemented at banks and regulatory institutions, the result would be an AI-based money laundering compliance system reducing backlog, identifying less false positives and more cases of real money laundering, and exposing suspicious transactions on cryptocurrency blockchains.

There are a few concerns about an increased reliance on AI systems, most of which span any industry: replacement of jobs, potential for data breaches, and other existential problems like possible racial profiling. Let’s address these three briefly in turn:

First, employment replacement by machines is a legitimate concern in some industries, particularly manufacturing. But most analysis done by private sector consulting firms has projected that increased use of AI will lead to an uptick in employment in a number of industries, including financial compliance. That’s because with more sophisticated outputs coming from better systems, banks will have more decisions to make about which clients and institutions to do business with, requiring—you guessed it—more employees to do analysis on risk. While jobs may look different, they won’t necessarily be replaced.

Second, the potential for data breaches is a real one at every institution but is also one that already exists. There isn’t much significance in reconfiguring data that is already obtained at financial institutions. It is vitally important that banks and regulators continue protecting such data. However, it is quite important to protect the AI systems themselves from breach in order to protect algorithms from falling into the hands of those who would study them in order to learn how to avoid detection. Recent breaches like the FireEye breach—suspected to be connected to the ongoing cyberattack on U.S. federal agencies including the Treasury, Homeland Security, and Commerce Departments—demonstrate that underlying technology is just as critical to protect as underlying personal data. AI systems are no different and will need significant cybersecurity.

Finally, the issue of racial or ethnic profiling is a salient one in AI systems and money laundering compliance. Recent court cases determining that Muslim men have standing to sue the U.S. government over their inclusion on a “no fly” list as a result of their religion or ethnicity suggest that systems which build in racial or religious bias as an inherently suspect characteristic could face serious legal challenges going forward.

Of course, no one should be considered suspicious in any way simply because of their race, religion or any other suspect classification. Consequently, despite the fact that most activity in this space will remain private and unseen by the public, companies must still consider not just the law, but also ethical values in implementing the new systems. It is imperative to ensure improper biases do not cloud or distort the results.   

Biography

Robert J. DeNault (J.D. 2021) is a third-year law student and the Class of 1986 scholar at Duke University School of Law.  He was born and raised in Narberth, Pennsylvania and received a B.A. in Political Science from Fordham University cum laude before starting at Duke Law.  At Duke, he has focused on national security issues like cryptocurrency, money laundering, cyber intrusions and ransomware attacks.  He also writes for Duke’s Journal of Constitutional Law and Public Policy and plans to join Gibson Dunn & Crutcher as an associate after graduation.

The views expressed by guest authors do not necessarily reflect the views of the Center on Law, Ethics and National Security, or Duke University.

Remember what we like to say on Lawfire^®: gather the facts, examine the law, evaluate the arguments – and then decide for yourself!