Automated Adverse Media Screening with LLMs

A new approach to adverse media screening in Anti-Money Laundering (AML) and Know Your Customer (KYC) compliance processes is based on the use of Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG). Traditionally, these processes rely on keyword-based searches, which generate a high number of false positives or require extensive manual reviews.

The proposed system implements an LLM agent that performs web searches, retrieves and processes relevant documents, and calculates an Adverse Media Index (AMI) for each subject. This multi-step approach aims to improve the efficiency and accuracy of screening.

The system's evaluation, conducted using different LLMs, demonstrated its ability to distinguish between high-risk individuals (such as politically exposed persons, subjects on regulatory watchlists, and sanctioned persons) and low-risk individuals. The data used for testing comes from sources such as OpenSanctions and academic publications.