Papers for

financial crime analysts

Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.

CryptoFlowBench evaluates tracing illicit cryptocurrency fund flows

CryptoFlowBench: Benchmarking Multi-Hop Reasoning on Blockchains for Cryptocurrency Fund Tracing

Abstract: Pseudonymous accounts and complex smart-contract ecosystems substantially hinder the tracing of illicit cryptocurrency fund flows. Existing Graph Neural Networks (GNNs) have achieved promising performance in blockchain fraud detection, yet their static node- or edge-level predictions do not naturally generalize to multi-hop fund tracing. Meanwhile, existing tracing methods mainly rely on predefined heuristics to capture multi-hop relationships, limiting their adaptive reasoning capabilities. To bridge this gap, we introduce CryptoFlowBench, a benchmark constructed from authoritative reports on cryptocurrency-related illicit incidents, with four tasks spanning local to multi-hop fund tracing. These tasks pose substantial challenges to existing learning- and heuristic-based methods, particularly in complex multi-hop scenarios. We therefore propose ChainSleuth, a graph-tool-augmented agent that iteratively reasons, invokes graph tools, and updates its tracing trajectory, enabling adaptive information acquisition and recovery from erroneous paths. Experiments show that no baseline consistently dominates all four tasks, revealing substantial gaps between local fraud detection and multi-hop fund tracing. ChainSleuth achieves notable improvements on multi-hop tasks, demonstrating the potential of agentic graph reasoning for cryptocurrency fund investigation.

Mon 12 OctCryptography and SecurityArtificial IntelligenceMachine Learning
The gist
Tracking illegal money in cryptocurrencies is hard because accounts are anonymous and transactions are complicated. Existing methods either look at simple connections or use fixed rules, making it tough to follow money through many steps. To improve this, the authors created CryptoFlowBench, a set of tests based on real crimes involving cryptocurrencies, focusing on tracing funds across multiple transaction steps. They also developed ChainSleuth, a tool that explores transaction networks step-by-step and corrects its path when it goes wrong, showing better results at multi-step tracing than previous methods.
Open → 2610.16449v1

Feed-guided shortest paths improve illicit subgraph extraction in transaction networks

Extracting Illicit-Rich Subgraph from Transaction Networks via Feed-Guided Shortest Paths

Abstract: We study investigative subgraph extraction for Anti-Money Laundering (AML) transaction graphs using the Shortest Paths Network Search Algorithm (SPNSA), which aggregates shortest paths among a small input feed of entities. Since SPNSA is highly sensitive to the feed choice, we evaluate several graph-only feed selection strategies and propose a motif-based coherent feed selection method. Experiments on the IBM-AML benchmark (HI-Small, LI-Small, HI-Medium) and the Elliptic Bitcoin transaction graph show that the proposed strategy yields markedly higher illicit ratios in the extracted subgraph than random and degree-based heuristics on Elliptic and the IBM-AML HI variants, while keeping the output compact.

Fri 9 OctSocial and Information NetworksComputational Engineering, Finance, and ScienceComputers and Society
The gist
Finding suspicious patterns in money transaction networks helps fight crime like money laundering. This paper looks at a method that extracts important parts of a network by finding shortest paths between some starting points called a feed. The authors show that picking these starting points carefully, using a new motif-based approach, leads to finding more illicit activity than random or simple methods. Their tests on real and benchmark data confirm their approach finds clearer suspicious subgraphs while keeping results manageable in size.
Open → 2610.13280v1

Interpretable method detects money laundering in Ethereum transactions

An Interpretable Approach to Money Laundering Detection in Transaction Graphs using Pass-Through Templates

Abstract: Layering is a key stage of the money laundering process in which assets are moved through intermediate entities to obscure their origin. We model this movement as a transaction graph, a labeled directed multigraph where nodes represent entities, such as individuals or organizations, and edges represent transactions between them. This work addresses the detection of layering patterns in transaction graphs. We define pass-through templates, a class of transaction graphs whose structure is indicative of a specific layering pattern in which entities receive funds and rapidly forward them. We formulate detecting instances of these templates within a larger transaction graph as a maximum-cardinality, maximum-weight matching problem. The resulting method uses Edmonds' blossom algorithm, which provably finds an optimal solution to this problem. Here we present a case study on publicly available Ethereum transaction data. We identify 283 pass-through template instances, chained through shared address nodes into a densely interconnected network that moves hundreds of millions of dollars in stablecoin value. We find that six addresses in the identified network are sanctioned by the U.S. Department of the Treasury for associations with criminal organizations, providing evidence of ties between the network and sanctioned criminal entities.

Thu 17 SeptSocial and Information Networks
The gist
Money laundering tries to hide where money originally comes from by moving it quickly through many accounts. The authors created a clear way to spot this behavior in networks of transactions by looking for patterns where money is passed through several entities rapidly. They use a mathematical approach to find these patterns optimally and tested it on Ethereum blockchain data. Their method found hundreds of suspicious patterns linked to millions of dollars and some accounts already known for crime.
Open → 2609.20737v1