Back in 2019, blockchain analytics firm Elliptic published research with the MIT-IBM Watson AI Lab showing how a machine learning model could be trained to identify Bitcoin transactions made by illicit actors, such as ransomware groups or darknet marketplaces.

Now the partners have put out new research applying new techniques to a much larger dataset, containing nearly 200 million transactions. Rather than identifying transactions made by illicit actors, a machine learning model was trained to identify “subgraphs”, chains of transactions that represent bitcoin being laundered.

Identifying these subgraphs rather than illicit wallets let the researchers focus on the “multi-hop” laundering process more generally rather than the on-chain behaviour of specific illicit actors.

Working with a crypto exchange, the researchers tested their technique: of 52 money laundering subgraphs predicted and which ended with deposits to the exchange, 14 were received by users who had already been flagged as being linked to money laundering.

On average, less than one in 10,000 accounts are flagged in this way "suggesting that the model performs very well," say the team. The researchers are now making their underlying data publicly available.

Says Elliptic: "This novel work demonstrates that AI methods can be applied to blockchain data to identify illicit wallets and money laundering patterns, which were previously hidden from view.

"This is made possible by the inherent transparency of blockchains and demonstrates that cryptoassets, far from being a haven for criminals, are far more amenable to AI-based financial crime detection than traditional financial assets."