Tracing illicit cryptocurrency flows is notoriously difficult. Pseudonymous accounts and complex smart-contract ecosystems obscure the movement of funds, making it hard to identify criminal activity. A new paper on arXiv introduces CryptoFlowBench, a benchmark designed to evaluate multi-hop reasoning on blockchains for cryptocurrency fund tracing.
The benchmark specifically targets the challenge of following funds across multiple transactions and interactions, rather than just single-hop transfers. This is a key requirement for tracing laundered or stolen assets in decentralized finance. The authors note that existing Graph Neural Networks (GNNs) have shown promising performance in blockchain fraud detection, suggesting that CryptoFlowBench could provide a standardized way to test and compare such models.
By focusing on multi-hop reasoning, CryptoFlowBench aims to push forward research on more robust tracing methods. The paper is available as arXiv:2610.16449.