Towards Explainable and Effective Anti-Money Laundering for Cryptocurrency

Cryptocurrency money laundering poses a serious threat to the security and compliance of decentralized financial systems. While various detection methods have been proposed, existing approaches either lack explainability or struggle to effectively handle privacy coins. This dissertation proposes two research directions aimed at enhancing the explainability and effectiveness of anti-money laundering (AML) techniques. The first direction proposes an explainable AML framework for non-privacy coins by incorporating transaction semantic parsing and large language model-based classification. The second direction explores a novel method for tracking illicit fund flows involving privacy coins such as Monero, focusing on improving de-anonymization accuracy and identifying cross-chain laundering paths. Preliminary results on the first direction demonstrate the feasibility of a semantic-aware module and the potential of a fine-tuned large language model in detecting complex laundering behaviors.