A Defamation-Safe, Evidence-Anchored Communication Protocol Between AI Agents in Workers’ Compensation SIU Workflows

Abstract

Artificial intelligence is increasingly deployed in workers’ compensation claims operations to surface inconsistencies, anomalies, and patterns warranting Special Investigations Unit review. However, many AI-enabled fraud detection systems fail under regulatory examination, discovery, or litigation—not due to model inaccuracy, but because their outputs are not governable, reproducible, or legally defensible. Unstructured communication between automated systems can embed implicit accusations, omit evidence provenance, and obscure accountability for decisions affecting statutory benefits. This paper introduces a formal, defamation-safe, evidence-anchored communication protocol governing interactions between two AI agents: a Fraud Recommendation Agent and an SIU Triage Agent. Designed for workers’ compensation claims, the protocol defines message contracts, evidence bundle requirements, language constraints, decision boundaries, and human-in-the-loop escalation points. Rather than automating fraud adjudication, the protocol establishes a systems-level governance framework preserving investigative integrity while enabling continuous learning. The contribution is a citable, enterprise-grade design pattern reframing AI-enabled SIU workflows as a coherence and accountability problem rather than a prediction accuracy problem.