The MIT-IBM Watson AI Lab and blockchain analytics company Elliptic collaborated to publish research in 2019 that demonstrated how a machine learning model could be trained to recognize Bitcoin transactions conducted by bad actors, such ransomware organizations or darknet marketplaces.

The partners have now published fresh research that uses novel approaches on a considerably larger dataset—nearly 200 million transactions—in their work. A machine learning model was taught to detect “subgraphs,” or sequences of events that indicate bitcoin being laundered, as opposed to transactions done by illegal actors.

The researchers were able to concentrate on the "multi-hop" laundering process more broadly as opposed to the on-chain actions of particular illicit actors by identifying these subgraphs rather than criminal wallets.

The researchers evaluated their method using a cryptocurrency exchange: out of 52 money laundering subgraphs that were predicted and resulted in deposits to the exchange, 14 were received by individuals who had previously been reported as having a connection to money laundering.

This kind of flagging occurs in less than one in 10,000 accounts on average, "suggesting that the model performs very well," the team reports. The researchers are now opening access to the underlying data to the general public.

Elliptic states: "This innovative work shows how artificial intelligence techniques may be used on blockchain data to find hidden patterns of money laundering and unlawful wallets.

 

The inherent transparency of blockchains makes this possible and illustrates that cryptoassets, contrary to the perception of being a haven for criminals, are more conducive to AI-based financial crime detection than traditional financial assets.