AI-powered prior authorization pilots: what builders need to know about Medicare's experiment

The federal government is piloting AI-powered prior authorization for Medicare, testing whether artificial intelligence can determine eligibility and coverage levels for insured individuals. For AI builders in healthcare, this pilot signals where payer workflows are heading and what fairness, transparency, and auditability requirements may follow.

What happened

The government is piloting a program that uses AI to make insurance coverage decisions, specifically for traditional Medicare. The pilot is being tested across six states: Arizona, Ohio, Oklahoma, New Jersey, Texas, and Washington. The goal is to evaluate whether AI can expedite approval of unambiguously allowable claims and reduce care delays. However, the pilot has drawn opposition. Representative Greg Landsman (D-Ohio) co-sponsored the Ban AI Denials in Medicare Act, arguing the pilot should be stopped.

Why AI builders should care

This pilot is one of the first government-led efforts to embed AI directly into payer coverage decisions. For teams building AI products for healthcare administration, the pilot creates a test case for how AI systems handle high-stakes decisions with regulatory scrutiny. The outcomes could influence future requirements for explainability, appeals processes, and bias auditing in automated PA systems. A nationwide survey found that roughly 3 out of every 4 health plans already use AI for prior authorization approvals, so the Medicare pilot is part of a broader industry shift.

Practical implications

Proponents argue that AI could speed up approvals for routine claims, reducing administrative delays for patients and physicians. But the AMA reports that over 60% of doctors say unregulated AI tools systematically deny patients coverage for necessary care. Critics worry that AI-driven prior authorization may increase wrongful denials and obscure decision logic, making it harder for patients and clinicians to understand or appeal denials. Some experts predict an escalation: better AI denials will be met with better AI appeals, increasing volume without fixing the underlying misalignment.

For AI builders, the practical takeaway is that any system deployed in this space will need robust audit trails, clear decision rationale, and mechanisms for human review. The pilot's design and results will likely inform future regulation around AI in insurance claims automation.

Caveats

The available evidence focuses on pilot concepts and policy debates rather than concrete agency details, timelines, or specific AI models. The exact scope of the pilot, the AI systems being used, and the evaluation criteria are not consistently documented across sources. Builders should monitor official announcements from CMS for technical specifications and compliance requirements.