Banks are testing AI to catch money laundering faster — and it's working better than their old rule-based systems

Researchers built a machine learning system that spots suspicious financial transactions more accurately than the rule-based systems banks currently use. Banks process billions of transactions daily and drown in false alarms; this system reduces false positives while catching more actual suspicious activity, which means fewer wasted investigator hours.

Banks have relied on the same rigid, rule-based detection approach for decades — flagging transactions that match predefined patterns. The problem: those rules generate so many false alarms that investigators get buried and real money laundering slips through. This paper shows that a simpler machine learning approach outperforms those old systems on real bank data, which is the kind of evidence that actually moves institutions toward deployment. If banks start adopting this or similar ML-based detection, it means faster identification of criminal money flows and fewer resources wasted investigating innocent transactions. The catch: this is a research paper, not a deployed system, so the real question is whether banks will actually integrate it into their operations or stick with what they know.