Blockchain is often criticised as a solution in search of a problem. But one group of people has already found immense value in the tech: money launderers.
Their crimes cause painful headaches for financial institutions, crypto businesses, legitimate law enforcement agencies, and cryptocurrency regulators. All of them need to tackle illicit activity on blockchains.
Enter Elliptic, a British firm that specialises in cryptocurrency forensics. The company uses blockchain analytics to protect customers from financial crimes. Its latest safeguard taps the power of AI. By applying machine learning to transaction data, Elliptic has created a new method of detecting money laundering on the Bitcoin blockchain.
New research reveals the technique has already identified proceeds of crime sent to a crypto exchange, novel patterns of money laundering, and previously-unknown illicit actors.
The outputs have already been baked into Elliptic’s products. With further testing, the company believes the AI model could directly flag illicit transactions.
Tom Robinson, Elliptic’s co-founder and chief scientist, envisions two core use cases.
“The primary application is in anti-money laundering, helping crypto exchanges and other businesses to identify crypto that may have originated from criminal activity,” Robinson told TNW via email.
“It could also be used by law enforcement agencies to identify new illicit services and actors making use of cryptocurrencies.”
A model approach to money laundering
As decentralised and (pseudo)anonymous transaction systems, blockchains are extremely attractive to money launderers. Unfortunately for them, blockchains are also ripe for AI analysis.
By scanning ledgers of transactions and data on wallets, machine learning can spot signs of illicit payments — and the criminals behind them.
“This probably makes cryptocurrencies more amenable to AI-based financial crime detection than traditional financial assets,” Robinson said.
Elliptic has explored the possibilities for years. Back in 2019, the company developed a machine learning model that found Bitcoin transactions made by illicit actors, such as ransomware groups and darknet marketplaces.
The new research updates the techniques and applies them to an enormous dataset, containing over 200 million transactions.
Elliptic worked with researchers from the MIT-IBM Watson AI Lab to develop the method. They decided to apply a novel approach.
Rather than focusing on illicit wallets, the researchers trained their model on “subgraphs,” which represent bitcoin transaction chains. Some of these transactions involved money laundering.
By focusing on subgraphs rather than wallets, the model could analyse the broader “multi-hop” laundering process.
Identifying money laundering with AI
After training, the team applied their techniques to real transactions on an unnamed cryptocurrency exchange. Their model promptly identified 52 suspicious subgraphs that ended with deposits. The findings were then sent to the exchange for review.
The response was promising. Fourteen of the 52 accounts that received the deposits had already been flagged for links to money laundering — and the real number could be higher.
Although the remaining 38 accounts had no definitive connection to illicit activity, that doesn’t mean they’re guiltless.
The exchange had flagged under 0.1% of its customer accounts as suspicious. Yet these insights were based entirely on off-chain information.
The model conducts a deeper analysis. Adding this capacity to the confirmed predictions suggests the system is already performing well. With further refinement, Elliptic hopes to commercialise the techniques.
The company also investigated the laundering patterns that the model spotted. These included “peeling chains,” which wash large quantities of cryptocurrency through a lengthy series of small transactions, and “nested services,” which use larger exchanges to conceal illicit trades.
By identifying these behaviours, Elliptic plans to expand the scope of its tools. The company has also made the underlying data publicly available.
“We’re only scratching the surface of what is possible here,” Robinson said. “This work will be extended to other blockchains and will become more and more effective as we incorporate more data.”