A new AI model developed by MIT-IBM and Elliptic analyzed 200,000 Bitcoin transactions, successfully detecting 52 cases of money laundering on a cryptocurrency exchange.
Scientists from AI MIT-IBM Watson Labs and the world's leading blockchain analysis company Elliptic have successfully researched and developed a new artificial intelligence (AI) model, capable of detecting money laundering transactions. Bitcoin is based on over 200,000 analyzed transactions and has an impressive accuracy rate. The main goal of the research is to explore the potential of AI in improving the detection mechanism of financial crime schemes on blockchain networks.
Elliptic's new research, published on arXiv, marks a significant step forward in applying artificial intelligence (AI) to combat money laundering in the cryptocurrency market. The difference of this study lies in focusing on the "subgraph", which is the transaction chain that represents the entire Bitcoin money laundering process. Instead of just identifying individual illicit wallets, Elliptic's AI model is capable of tracking the flow of illicit funds through multiple intermediate steps, helping to expose complex money laundering networks.
The training data contains “subgraphs”: sequences of transactions, some of which are known to represent money laundering activity.
To verify its real-world effectiveness, Elliptic tested its AI model with a major cryptocurrency exchange. The results showed that AI correctly identified 14 out of 52 money laundering cases, equivalent to an accuracy rate of nearly 27%. It is worth noting that these cases were confirmed based on “off-chain” information, meaning data not available on the blockchain, proving that AI is capable of outperforming traditional blockchain analysis techniques.
Besides detecting known money laundering models such as “peeling chains”, Elliptic's AI model can also identify new methods such as the use of “Nested Services – (sophisticated money laundering method, leveraging many intermediary service layer to hide the source of money). This knowledge is extremely valuable for AML (anti-money laundering) professionals and can be integrated into Elliptic tools to enhance detection of illegal activities.
Simple illustration of two examples of money laundering patterns identified by the AI model.
Elliptic also announced the “Elliptic2” dataset, a public dataset containing more than 200 million Bitcoin transactions, allowing the research community to access and develop new techniques to combat money laundering more effectively.
This research is a strong testament to the potential of AI in the fight against financial crime in the cryptocurrency sector. Although illicit cryptocurrency trading volumes decreased in 2023, more than $24.2 billion in assets remained tied up in illegal activities, of which stablecoins accounted for more than 60%.
