MASTER Assignment
Privacy-preserving detection of dense money-laundering flows across institutions
Type : Master M-CS
Period: December 2025 - June, 2026
Student: Kruk, N. (Niels, Student M-CS)
Date Final project: June 12, 2026
Supervisors:
Abstract:
Effective anti-money-laundering (AML) detection requires analysing transaction networks that span multiple financial institutions, yet privacy regulations prevent banks from sharing raw transaction data. Existing detection methods assume a single institution with full visibility over its transaction graph and are therefore unable to identify laundering patterns that cross institutional boundaries.
FlowScope is a flow-based graph algorithm for detecting money-laundering patterns in transaction networks by extracting subgraphs in which there are high-volume, multi-step flows from source accounts through low-retention intermediary accounts to destination accounts. This thesis presents a privacy-preserving protocol that adapts the FlowScope algorithm to a multi-institution setting using secure multiparty computation (MPC). The protocol realises the FlowScope greedy-peeling procedure across three semi-honest compute servers using secret sharing and oblivious RAM. We also reformulate FlowScope’s source and destination account definitions to remain semantically well-defined when the transaction graph is distributed across multiple banks.
We provide a formal protocol specification, a simulation-based security proof under the semi-honest model, and an empirical evaluation that verifies correctness against a plaintext baseline and measures communication and computational complexity. This is the first MPC protocol for fast-flow subgraph detection in a distributed financial setting.