Anti-money laundering (AML) in financial networks depends on detecting transaction patterns that span multiple institutions. In practice, however, banks cannot share account and transaction features due to strict privacy regulations, which limits the usefulness of graph-based machine learning models.

A new paper on arXiv proposes a solution using homomorphic encryption, a technique that allows computation on encrypted data. The authors describe a framework for privacy-preserving GNN inference on financial transaction graphs, where the model can operate on encrypted features without revealing the underlying data.

The work addresses a key tension between regulatory compliance and collaborative AML efforts. By enabling inference on encrypted data, the approach could allow banks to benefit from cross-institutional graph models while keeping sensitive information private. As this is a single source, no conflicting findings are reported.