Exploring the In-Context Learning Capabilities of LLMs for Money Laundering Detection in Financial Graphs

TL;DR

Detecting money laundering in graph-structured financial data and producing explanations for decisions is addressed. The authors propose a three-stage pipeline that extracts $k$-hop subgraphs, serializes them to text, and uses few-shot prompting with AML typologies to query an LLM for suspiciousness and rationale. They demonstrate empirical feasibility on synthetic IBM AML data, showing that LLMs can highlight red flags and provide human-readable justifications, enabling explainable graph-based AML analytics. The work motivates hybrid, language-driven fraud analytics and provides a foundation for scalable, interpretable systems that combine fast graph methods with LLM reasoning.

Abstract

The complexity and interconnectivity of entities involved in money laundering demand investigative reasoning over graph-structured data. This paper explores the use of large language models (LLMs) as reasoning engines over localized subgraphs extracted from a financial knowledge graph. We propose a lightweight pipeline that retrieves k-hop neighborhoods around entities of interest, serializes them into structured text, and prompts an LLM via few-shot in-context learning to assess suspiciousness and generate justifications. Using synthetic anti-money laundering (AML) scenarios that reflect common laundering behaviors, we show that LLMs can emulate analyst-style logic, highlight red flags, and provide coherent explanations. While this study is exploratory, it illustrates the potential of LLM-based graph reasoning in AML and lays groundwork for explainable, language-driven financial crime analytics.

Figures (3)

• Figure 1: Pipeline Overview for LLM-based AML Analysis.
• Figure 2: Snapshot of a sample 2-hop subgraph centered around a transaction between acct_80FF89190 and acct_810147BB0.
• Figure 3: Canonical money laundering patterns used as reasoning demonstrations. Image adapted from the IBM AML dataset paper.