Will AI Fix Prior Authorization—Or Make It Worse?
Prior authorization is one of the most friction-laden processes in American healthcare. Physicians spend hours each week submitting requests to insurers who must approve treatments, medications, or procedures before they can proceed. Delays routinely result in abandoned care, worsened patient outcomes, and administrative burnout. The process was designed as a cost-control mechanism; in practice, it has become a significant operational burden on the clinical workforce.
AI is now entering this process from both sides. Hospitals and physician groups are deploying AI tools to generate and submit prior authorization requests faster. Insurers are deploying AI to review and adjudicate those same requests. The collision of these two trends is what makes the current moment worth examining closely.
On the provider side, AI systems can pull from patient records, match clinical criteria, and draft authorization requests in a fraction of the time a human administrator would require. Tools from companies like Cohere Health and Waystar are positioned to reduce the clerical load on clinical staff and accelerate the submission pipeline. For health systems operating at scale, this represents a measurable reduction in labor costs and turnaround time.
On the payer side, insurers including UnitedHealth Group have faced public and legal scrutiny for using AI systems to deny claims at high volume with limited human review. The concern is not that AI is being used—it is that AI may be optimized to maximize denial rates within defensible criteria windows, effectively automating a decision that previously required human judgment to contest. When a system is trained on historical approval and denial patterns, it can reproduce those patterns with greater speed and consistency, including any embedded bias toward denial.
The regulatory environment has begun to respond. The Centers for Medicare and Medicaid Services finalized rules in 2024 requiring payers to provide specific denial reasons and faster turnaround times. Several states have moved to restrict fully automated denials without physician review for certain care categories. These interventions acknowledge that automation without accountability creates systemic risk for patients who lack the resources to appeal.
The deeper operational question is whether AI on both sides of the prior authorization process creates equilibrium or escalation. If providers use AI to submit more complete, criteria-aligned requests, and insurers use AI to process them faster with consistent logic, the theoretical outcome is a faster, lower-friction system. The realistic concern is an adversarial dynamic: provider-side AI learns to write requests that match payer-side AI approval patterns, while payer-side AI is periodically updated to close those pathways. The administrative burden does not disappear—it shifts to the teams managing, auditing, and retraining these systems.
For health systems and insurers evaluating AI deployment in this domain, the operational stakes are concrete. Faster authorizations reduce cost and improve care continuity. Automated denials without explainable criteria create legal exposure and erode provider relationships. The technology is sufficiently mature to handle the task; the governance layer—who reviews AI decisions, at what threshold, and with what appeal rights—is where the meaningful risk resides.
What the prior authorization case illustrates more broadly is that deploying AI into a structurally contested process does not neutralize the underlying conflict. It accelerates it. The efficiency gains are real, but so is the potential to embed and scale adversarial dynamics that were previously limited by human bandwidth. Organizations that treat AI deployment here as purely a throughput problem, without redesigning the accountability structures around it, are likely to create new categories of operational and legal exposure rather than resolve the old ones.
Sources: — Ars Technica (https://arstechnica.com/ai/2026/07/will-ai-fix-prior-authorization-or-make-it-worse/)