CLPS Completes AI Assisted Anti Money Laundering Project

CLPS Incorporation said in a press release that it completed an AI assisted anti money laundering review project for a Chinese commercial bank. The project took place within a Hong Kong regulatory department's Gen A.I. Sandbox and used a fine tuned, small open source language model to generate risk ratings.

The model achieved accuracy above 90%, compared with more than 40% for a general purpose model and just over 30% for a model with 400% more parameters. CLPS used synthetic data, risk distribution matching, class weighted loss functions and stratified sampling to improve results, including for high risk cases.

CLPS separated risk ratings, information requests and final reports into distinct outputs to reduce hallucinations and interference between tasks. It also used a seven category scorecard, blinded testing by another language model and human review, with the bank's specialists retaining final decisions for edge cases.