Explainable Aml Triage with Llms: Evidence Retrieval and Counterfactual Checks

Anti-money laundering (AML) transaction monitoring generates a high volume of alerts that need to be quickly reviewed by investigators under strict audit and governance constraints. Current methods often rely on rule-based systems, which can lead to high false positive rates and inefficient use of investigator time. A new study published on arXiv explores the potential of using large language models (LLMs) to improve AML triage. The researchers propose using LLMs to retrieve relevant evidence from unstructured data and conduct counterfactual checks to reduce false positives. This approach has the potential to improve the efficiency and effectiveness of AML investigations.

The study's authors argue that LLMs can be trained to identify relevant evidence and provide explanations for their decisions, enabling investigators to understand the reasoning behind the model's output. This can help to build trust in the model's recommendations and improve the overall quality of AML triage. The authors also suggest that the use of LLMs can help to reduce the workload of investigators by automating routine tasks and freeing them up to focus on high-risk cases.

The study's findings suggest that the proposed approach is effective in reducing false positives and improving the efficiency of AML triage. However, the authors also note that further research is needed to fully realize the potential of LLMs in AML triage and to address potential challenges such as data quality and model explainability.