Money Laundering Finally Meets Its Match – Federated Learning Will Change the Game

Money Laundering Finally Meets Its Match – Federated Learning Will Change the Game

Criminals move illicit funds through numerous financial intermediaries to avoid detection – that is the essence of money laundering. Money launderers are inventive in cleaning money. There are various money laundering approaches, but they all have something in common: at the end of the line is a financial institution trying to catch them, and that they are trying to fool. But things are changing, and modern technology promises to significantly help solve this problem. 

In 2021, The Financial Action Task Force (FATF) said “using collaborative analytics, financial institutions can better understand, assess, and mitigate money laundering and terrorist financing risks.” However, FATF also noted, “data pooling and collaborative analytics also have the potential to infringe on the protection of individual and fundamental rights to privacy.”  So, the ability to share insights and protect privacy comes with extraordinary challenges.

Toll on Society

While money laundering is often thought of as a “victimless crime,” it finances some of the world’s most malign industries, including terrorism, the drug trade, human trafficking and even wildlife smuggling. In fact, the U.N. Office on Drugs and Crime estimates that 2-5% of global gross domestic product (GDP) is laundered money, which amounts to $800 million to $2 trillion every year. And this is before we even begin to talk about the cost of anti-money laundering (AML) enforcement and the tolls that these bad industries take on society.

In more recent times, money laundering has become increasingly hard to police. This has been caused by our growing digital economy and the rise of cryptocurrency, whose value proposition is to cover the tracks of the person spending it (which is exactly what money launderers want). Even using more conventional means, money laundering is a widespread problem that usually ends in the same place: payments passing through and deposits being made into legitimate banks and other financial systems.

Challenges in the Way

Stopping money laundering, terrorist funding, human trafficking, etc. remains a top priority for regulators, who continue to ensure that financial institutions are properly focused on identifying the behaviors that are aligned to these crimes. Interestingly though, while the cost to organizations continues to skyrocket, many of the anti-money laundering (AML) tools and technologies remain quite old and outdated. Most were created decades ago and were never meant to cope with today’s world of ubiquitous access, clever criminals, constant mergers-and-acquisitions, and decentralized cryptocurrencies.

Alongside of this, privacy regulations are one of the biggest challenges faced by financial institutions. In Europe, this involves GDPR (General Data Protection Regulation) regulations, and in the United States, it is part of federal regulations and virtually every state’s privacy laws. These regulations make it immensely hard for banks to share information with each other to catch money launderers; it can also inhibit large banks with operations in different jurisdictions and countries cannot share information across borders, even though the data is internal and part of the same organization.

Additionally, the conspiracy of old technology, the risks and cost of change, clever criminals and a growing regulatory regime create a problem with which professionals in financial crime units and cybersecurity are all too familiar: the enormous amount of false positives. Old AML systems generate between 97-98% false positives, resulting in financial crime organizations spending most of their time chasing down could-be money laundering events that don’t amount to anything. 

Federated Learning Changes the Game

Some financial institutions have tried to address money laundering by centralizing customer data, and then running analytics on the data to look for patterns. Some jurisdictions have tried a form of sharing suspicious activity reports (SARs). But so far, very little has changed. 

Both approaches have their problems. Data consolidation or modelling projects are hugely expensive and there are data consistency challenges, particularly where data is pooled from different organizations (there are almost insurmountable privacy and security risks, that are time-consuming and laden with potential quality issues). 

The key is to find a way for financial institutions to share insights. To take learning from across the entire financial system and share it but leave the data where it is and protected. Where one institution sees a pattern of possible money-laundering behavior and shares it with peers – and to address privacy and security considerations – to do it without moving the data and protecting individuals’ and entities’ privacy. 

This is where federated learning (FL), a subset of machine learning, comes into its own. FL can share insights across financial institutions and within large organizations that have unique business units without ever moving data from its origin. It does this by moving the algorithm to the data, training it on that data, taking the learnings and improving the ability to detect suspicious activity. The important thing is the data never moves. Learning from disparate data is not new, but doing so and ensuring privacy in extremely sensitive and regulated environments is. It is why FL is so important in the fight against money laundering and associated crimes.

Sharing the Wealth of Information

By moving analytics to the data, FL models can draw upon unusual behaviors across the industry (like when a customer makes unusual ongoing large deposits, or when a company makes a deposit that far exceeds what one would expect). It also enables financial institutions to collaborate, so they can share experiences. As a result, organizations can identify risks that were previously hidden because they didn’t see them or have the knowledge and data to build effective solutions and models. In this way, FL benefits the entire system. 

Technology usually lags the techniques of the criminals and the emergence of regulations. Thankfully, the financial services industry is playing catch-up with FL, promising to dramatically reduce false positives so it can finally put a dent in the dirty money laundering industry.