Listen to Machine Learning for AML in the Cryptocurrency Age (8 min):

The emergence of cryptocurrency and networks like Bitcoin have sparked a cultural clash in the banking world. While traditional banks and financial institutions might want to avoid the risks that come with cryptocurrency - such as giving safe haven to money laundering activities - having the right machine learning models in place enables financial institutions to offer the new payment methods that their consumers increasingly expect, combat money laundering, and upgrade their anti-money laundering (AML) compliance efforts.

Machine learning technology can uncover patterns that would otherwise go unnoticed by humans, such as detecting how far removed a crypto wallet is from another account or finding wallets with ties to known criminal activity. Armed with this knowledge, banks can get a clearer sense of which transactions are high risk, and AML compliance teams can reduce the volume of false positives they encounter. 

AML Challenges Posed by Cryptocurrency

Cryptocurrency-based transactions present a big challenge for banks’ AML compliance teams. 

In a blockchain, an address is a unique identifier (similar to an email address or mobile phone number) that provides a location where funds can be received, stored, and sent. Typically, this unique address is a long alphanumeric string of 26 to 35 characters. It is also public and can be shared freely without security concerns. The ability to access the funds is contingent on knowing a private key.

Transfers are movement of funds from one address to another in the form of cryptocurrencies. One of the most interesting properties of cryptocurrencies is that all transactions are publicly visible to everyone.

A key distinction between these addresses and bank accounts is pseudo-anonymity. When a client opens a bank account, there is mandatory proof of identity and KYC/CDD standards that FIs must follow. This information links the financial transactions to an identity, such as a person or a company. 

However, there are no such requirements to have a cryptocurrency address, and this address can be used as a pseudonym for the owner’s identity. As a result, it takes a great deal of financial forensics to connect addresses to real-world identities. Success is far from guaranteed and often depends on security lapses from the address owner.

Hence, cryptocurrencies present opportunities and challenges for anti-money laundering efforts. On the one hand, all transactions between addresses are publicly available. Typically, this is not the case with fiat currencies, as banks have only partial visibility (i.e., incoming transfers, outgoing transfers, and transfers between internal accounts). On the other hand, there are no information guarantees about the real-world identities behind the addresses, including basic KYC/CDD requirements.

Actually, many AML regulations depend on the identity of the sender and/or the receiver of funds, such as sanctions screening or terrorist watchlists. When it comes to cryptocurrencies, the missing identity information leaves banks struggling to determine whether those involved in the transaction are trustworthy or not.

Often, investigations leverage open-source intelligence, collaborations with regulated crypto exchanges and financial service providers, and the dark web, including active dark markets and exposed associations with illicit activity (e.g., scams, ransomware, malware, or terrorism). Investigations, however,  are very complex, specialized, time-intensive, and as a result, very hard to scale.

How Machine Learning Detects Money Laundering in Cryptocurrency

Machine learning can help fill in the information gaps created by cryptocurrency’s pseudo-anonymity. This is particularly true for technology that includes the ability to extract information from a blockchain ledger and review the entire transaction history between public addresses.

Active learning, for example, can focus on optimizing banks’ investigations. With active learning in place, investigators receive feedback about particular transactions. This enables them to train a machine learning model to detect suspicious patterns with few labels. 

Another example illustrates some of the challenges and opportunities of using machine learning to tackle money laundering in crypto. Consider a graph where nodes are cryptocurrency addresses, and edges represent transactions between these addresses. Machine learning with active learning investigates the “neighborhood” of a given address, including the average, minimum, and maximum distance to exposed illicit addresses, defined as the number of hops required to reach an illicit address. This information can then be conveyed to a machine learning model responsible for assessing the risk of money laundering. The most important assumption is that illicit actors have a degree of proximity or association between them.

In practice, consider address A, for which we have no information, due to pseudo-anonymity, except incoming and outgoing transactions. In conventional terms, it is tough to evaluate the money laundering risk of A, given that we know nothing about the identity of its owner. However, when we consider all visible transactions, we understand that some of A’s direct connections are connected to many exposed illicit addresses. Some machine learning techniques are designed specifically for graphs (such as the graph of transactions in a payment network). These techniques can produce information that summarizes the transactions between cryptocurrency addresses and pass that summary information downstream to a machine learning model that assesses risk or detects suspicious activity. 

In this respect, machine learning leverages one of the novel things about crypto: the availability of all transactions in public ledger. The technology can enlarge banks’ AML compliance capabilities to cryptocurrency use cases and detect patterns that might be connected to suspicious activity. 

For Banks, a Roadmap to Legitimate Crypto Use Cases

Trust is essential for banks as they consider investing in cryptocurrency-based solutions that a growing share of bank customers want to use. A recent survey of nearly 4,000 U.S. bank customers found 60% of them want banks to invest in cryptocurrency. The survey also found 68% of surveyed crypto owners are interested in Bitcoin-based credit or debit card rewards. 

However, the same survey also found banks are not ready to take the leap into crypto yet. Only 2% of surveyed bank executives said they were very interested in offering cryptocurrency, while 19% said they were somewhat interested. 

Banks are risk-averse by nature, so these results are understandable. But a cultural shift is underway and gaining strength. As bank customers turn to cryptocurrency tools, banks can’t avoid the culture clash any longer. But the good news is that cryptocurrency doesn’t have to be such a risk. Machine learning can give banks the insights and knowledge to implement crypto-based solutions and prevent crimes like money laundering.   

What is RiskOps, and how can it help banks take more risks without compromising trust? Watch our on-demand webinar The Evolution of Risk and AML to learn more.