Complex money laundering schemes increasingly exploit the intersection of traditional finance and cryptoassets, creating new challenges for Financial Intelligence Units (FIUs) that rely on manual analysis of Suspicious Transaction Reports (STRs). Current detection methods are limited to single-entity data analysis, failing to identify cross-service patterns that characterize modern laundering networks. This PhD project aims to develop computational methods for early detection of complex money laundering by modeling the behavior of individual agents within these networks.

The research pursues three specific objectives: (1) identifying indicators of complex money laundering schemes integrating cryptoassets and fiat currency through literature review and stakeholder interviews; (2) modeling money laundering networks and analyzing behavioral patterns of individual agents using graph data structures and temporal analysis; and (3) developing an early detection model that aligns agent behavior with transactional data using machine learning techniques including Bayesian networks, graph neural networks, and large language models for STR narrative analysis.

Expected results include a validated computational model capable of cross-referencing STRs from heterogeneous sources to identify network relationships and predict money laundering activity at early stages. The impact extends beyond technical advancement to support law enforcement asset recovery operations, with potential applications across the 182 FIUs within the Egmont Group, ultimately improving detection rates from the current estimated 2% of criminal proceeds seized annually.