This memo proposes a national regulatory framework to govern the use of artificial intelligence in health insurance coverage determinations, with a focus on utilization management and prior authorization. Insurers increasingly deploy predictive tools to estimate recovery time, length of stay, and medical necessity, yet the logic, criteria, and limitations of these systems are often opaque to patients and clinicians. Drawing on evidence of high-volume prior authorization use, documented error and reversal rates, and appeal timelines that frequently outlast clinically relevant windows, the memo argues that existing oversight is fragmented across plan types and agencies and is not calibrated to the speed and scale of AI-influenced denials. Building on California’s SB 1120 as a model, the paper advances a care-centered “AI in Healthcare Bill of Rights” organized around three enforceable pillars: (1) transparent disclosure whenever AI influences an adverse determination; (2) documented accountability by a licensed clinician for any denial, delay, reduction, or termination of coverage relative to the treating clinician’s request; and (3) independent audit and validation of both tools and decision workflows, including audit-ready logging and disparity monitoring. The memo recommends an ERISA amendment to establish a uniform federal floor for these requirements, complemented by coordinated federal and state implementation and enforcement. It concludes that targeted guardrails can preserve the legitimate administrative benefits of predictive tools while protecting patient autonomy, reducing inappropriate denials, and restoring trust in coverage decision-making. This paper was submitted to Dr. Cynthia Bailey's CS121 Equity and Governance for Artificial Intelligence.
Patrick et al. (Mon,) studied this question.