A Rapidly Evolving Threat Landscape
The Limits of Static, Rule-Based Systems
The core challenge lies in the limitations of static, rule-based monitoring systems. Financial institutions require a faster, more accurate, and scalable approach that can adapt to emerging threats.
Current systems often struggle with a high rate of False Negatives (missed laundering cases) or an overwhelming volume of False Positives, creating a massive administrative burden for investigation teams.
There is a critical need for an intelligent layer capable of real-time reasoning and analysis — while maintaining full auditability and human-in-the-loop control.
What This Research Set Out to Prove
The primary objective was to evaluate and compare the effectiveness of four different modeling approaches for detecting suspicious financial activities, with a specific focus on improving detection power (Recall) and reducing missed cases.
Four Models. One Comprehensive Dataset.
The research utilized a comprehensive dataset containing over 9.5 million transactions. Four distinct processes were developed and tested:
The AI agents were built using the UiPath Cloud interface and leveraged Large Language Models (GPT-4o) to perform contextual analysis and prioritize red flags.
The Stack Behind the Intelligence
AI Agents Deliver Statistically Significant Gains
The integration of AI agents delivered statistically significant improvements in the detection of money laundering activities compared to standalone models.
Recall Performance by Model — Comparative Results
Rule-Based Model
Rule-Based AI Agent
CatBoost Model
ML-Based AI Agent
+8.7pp
Recall gain for the Rule-Based model. Adding an AI agent layer pushed detection from 65.3% to 74.0% — without replacing the underlying system.
93.7%
Peak performance via the ML-Based AI Agent. A 7.8 percentage point improvement over the standalone CatBoost model, detecting 9 in 10 laundering cases.
↓FN
Dramatic reduction in False Negatives. Agents identified complex patterns typically overlooked by base models — the cases most likely to cause real harm.
Scalable
Enhanced scalability and intelligence. AI agents interpret policies and analyze transaction context faster than manual review teams, enabling institutions to grow without proportional compliance headcount.
Overall, Cybiant’s optimal solution — an AI agent paired with a benchmark machine learning model — detected 93.7% of all money laundering cases while keeping a moderate False Positive Rate. Pairing this with a second AI agent trained on reducing False Positives can further improve AML-process efficiency.
Read the Full Study
The complete paper covers our full methodology, statistical analysis, model architectures, and extended findings — including a detailed breakdown of False Positive trade-offs across all four approaches.
Download the Research Paper
Free access. We may occasionally share relevant research from Cybiant — unsubscribe any time.
