Red-Teaming for Health Care Fraud Detection: A RAG-Based AI Approach

Health care fraud detection remains a critical challenge, particularly within Medicare and Medicaid systems. Traditional detection methods often fail when confronted with adversarial tactics, highlighting the need for more adaptive and proactive approaches. This manuscript leverages retrieval-augmented generation for red-teaming in the context of health care fraud detection. The proposed system generates diverse fraud scenarios that allow for stress-testing. We present the conceptual design, illustrate red-teaming using Medicare data, and discuss its integration with Bayesian decision-making frame-works. By embedding domain-specific contextual knowledge, the framework could enhance fraud identification and compliance analysis.