A new analysis from the National Health Law Program finds that roughly three out of every four health insurance plans now report using AI for prior authorization approvals, while a smaller but still significant share — an estimated 8% to 12% of plans — use AI specifically to support denial decisions, a distinction advocates say deserves far more regulatory scrutiny than it currently receives.
Why the Approval-Denial Distinction Matters
Using AI to speed up approvals of routine, clearly medically necessary requests is widely viewed as relatively low-risk and beneficial — it reduces administrative burden on both insurers and providers without denying anyone care. Using AI to support denial decisions carries fundamentally different stakes, since an erroneous denial can delay or block necessary care entirely, and the National Health Law Program’s analysis suggests this higher-risk use case, while less common, is far from rare.
The Regulatory Gap
Federal AI policy currently treats these two use cases largely the same way, according to the analysis, which argues this uniform approach threatens to undermine prior authorization reform efforts specifically because it fails to distinguish between AI accelerating approvals — a goal nearly everyone supports — and AI contributing to denials, where the consequences of errors fall directly and often severely on patients awaiting care.
Connecting to the Medicare WISeR Controversy
The findings add context to the ongoing controversy over CMS’s WISeR pilot, a Medicare program using AI-assisted prior authorization review in six states, which has faced intense scrutiny after FOIA-obtained documents revealed a vendor warned the system wasn’t ready before launch and that thousands of requests were denied over a three-month period, with some patients waiting as long as 83 days for resolution. The National Health Law Program’s broader industry findings suggest WISeR’s problems may reflect a more widespread pattern across the private insurance market as well, not an isolated government pilot program’s unique failure.
Vendor Claims Versus Independent Evidence
AI fraud-detection and claims-processing vendors have published their own figures claiming significant efficiency gains — one vendor reports flagging 72% of unusual claims, while another claims five-to-tenfold return on investment for insurers — but these figures come from the vendors themselves rather than independent verification, and none directly address the denial-rate and patient-outcome questions the National Health Law Program’s analysis raises.
The Patient Advocate Perspective
Patient advocacy groups argue that even a relatively small percentage of plans using AI for denials represents a meaningful population of patients at risk, particularly because AI denial systems can operate at a speed and scale that makes it harder for patients and providers to catch and appeal errors before care is significantly delayed — echoing concerns raised by physicians nationally who have called for risk-based review and clear stop rules for AI systems involved in coverage decisions.
What’s Next
The National Health Law Program has called for federal AI policy to explicitly distinguish between AI-assisted approvals and AI-assisted denials, with stricter oversight requirements for the latter, though no such distinction currently exists in federal rulemaking. The debate is likely to intensify as congressional scrutiny of the Medicare WISeR pilot continues and as more private insurers expand their own AI-based claims review systems.
Photo: vjohns1580 / PIXABAY via Pixabay
