Advix Blog

AML with AI power

Compliance Technology
Elliptic, a blockchain analytics firm, has developed an AI model to fight money laundering. This model will help detect illicit financial activities in the cryptocurrency domain, particularly on the Bitcoin blockchain.

Elliptic's new AI model is based on a dataset of nearly 200 million Bitcoin transactions. This is much larger than the previous dataset of 200,000 transactions used in their 2019 study. The current dataset, named "Elliptic2," includes 122,000 labelled "subgraphs" that represent chains of transactions associated with money laundering activities. Elliptic's deep learning model is trained on a vast dataset to improve the accuracy and efficiency of detecting financial crimes in the crypto space.

The AI model can identify new and complex money laundering patterns, such as "peeling chains," where funds are sent to multiple addresses to obscure their origin. It also uncovers the use of "nested services," where illicit funds are processed through intermediary accounts at larger exchanges, often without the exchange's knowledge.

The practical applications of this AI model have already shown promising results. In collaboration with a cryptocurrency exchange, Elliptic's model identified multiple "money laundering" subgraphs, some of which were confirmed by the exchange's own off-chain information. This shows the model can spot illicit activities that might evade traditional blockchain analysis.

Elliptic has made a big step towards community engagement and innovation by making the "Elliptic2" dataset publicly available. This dataset is the largest of its kind and is expected to spur further research and development of AI techniques for detecting illicit cryptocurrency transactions. By sharing this data, Elliptic hopes to advance the fight against financial crime in the crypto industry.

The partnership between Elliptic and the MIT-IBM Watson AI Lab shows how blockchain and AI can work together. Their work helps Elliptic’s product offerings and helps fight financial crime.

Read the full research paper here and access the Elliptic2 dataset here.