
Share
Newly surfaced documents deepen concerns about a Medicare experiment that lets AI help decide which treatments get approved, raising fresh questions about oversight, accountability, and what patients actually know about who, or what, is reviewing their care.
For the roughly 68 million Americans who rely on Medicare, prior authorization can feel like a black box. A doctor recommends a treatment, and then a bureaucratic process decides whether the government will actually pay for it. That process can mean the difference between getting care promptly and waiting weeks, sometimes while a condition worsens. Now Medicare is testing whether artificial intelligence should play a bigger role in that decision-making, and new documents are raising uncomfortable questions about how transparent, and how accountable, that experiment really is.
The pilot program, run through the Centers for Medicare and Medicaid Services, explores using AI tools to help process prior authorization requests. The idea isn't new. Private insurers have leaned on algorithms for years to flag claims, predict costs, and streamline approvals. Think of it like a spam filter for health claims: a tool meant to sort the obvious "yes" and "no" decisions quickly so human reviewers can focus on the harder cases. In theory, that frees up time and resources. In practice, when the stakes involve someone's cancer treatment or hip replacement, the margin for error gets a lot less forgiving.
That's the tension at the heart of this story. Medicare serves some of the most medically vulnerable people in the country: seniors, people with disabilities, patients managing multiple chronic conditions. A denied or delayed authorization isn't an abstract inconvenience for these patients. It can mean a missed surgery window, a lapse in medication, or a hospital stay that could have been avoided. When an algorithm, rather than a person, has a hand in that decision, the public has a legitimate interest in knowing how it works, what data trains it, and who is accountable when it gets something wrong.
Documents obtained and reviewed by STAT add new layers of concern to a program that has already drawn skepticism from patient advocates and some members of Congress. The specifics of what the documents show remain behind a paywall in the original reporting, but the fact that they're surfacing at all signals a program still grappling with basic questions of design and disclosure, months after it launched.
That matters because trust in these systems is fragile and easily broken. Health policy researchers have long warned that algorithmic tools in insurance decisions can replicate or even amplify existing biases if they're trained on historical claims data that reflects unequal access to care. A model that learns from decades of coverage decisions could end up automating the very disparities regulators are supposed to be correcting. Without clear documentation about training data, error rates, and appeal mechanisms, it's difficult for outside experts, let alone patients, to know whether that risk is being managed responsibly.
There's also the question of government oversight capacity. CMS has a long track record of running demonstration projects and pilot programs, testing new payment and care models before scaling them nationally. That's generally a sound approach. It allows regulators to catch problems in a contained setting before they ripple out to tens of millions of beneficiaries. But a pilot program only works as a safeguard if it's genuinely transparent, if independent researchers and journalists can access enough information to evaluate it, and if the agency is willing to pause or redesign the program when problems surface.

The reporting suggests that bar hasn't been fully met yet. New documents prompting new questions, rather than new answers, is not a great sign for a program that directly touches patient care. It echoes a pattern seen elsewhere in health tech regulation: tools get deployed under the banner of efficiency and innovation, while the rules governing their use lag behind, often written retroactively after problems emerge rather than proactively to prevent them.
None of this means AI has no place in prior authorization. Used carefully, these tools could genuinely reduce the administrative burden that burns out clinicians and frustrates patients. Automating routine, low-risk approvals could free up human reviewers to spend more time on complex cases that genuinely need judgment. That's a legitimate and worthwhile goal, especially given how much of the health care workforce's time gets consumed by paperwork rather than patient care. The problem isn't the technology itself. It's whether the guardrails around it are strong enough to match the stakes.
Congress and patient advocacy groups have pushed for years for more transparency around prior authorization generally, AI-assisted or not. Denial rates, appeal success rates, and the criteria used to make decisions have often been difficult for outsiders to scrutinize, even in traditional, human-reviewed systems. Layering AI on top of an already opaque process risks compounding that problem rather than solving it, unless regulators insist on clear documentation and public reporting as a condition of deployment.
Medicare beneficiaries deserve to know how decisions about their care are being made, and whether the systems involved have been tested for fairness and accuracy before being rolled out at scale. This pilot program sits at the intersection of two powerful forces: the real promise of AI to cut administrative waste, and the real risk of automating decisions that affect people's health without adequate accountability.
Getting this balance right isn't just a technical challenge. It's a test of whether government agencies can keep pace with the technology they're deploying, and whether patients can trust that efficiency gains aren't coming at the expense of their care. The questions raised by these new documents deserve clear, public answers, not just from CMS, but from the companies building these tools and the lawmakers responsible for overseeing them. Until that happens, the millions of people depending on Medicare for timely, appropriate care are left wondering exactly what, or who, is making decisions about their health.
Tags
Original Sources
New documents raise more questions about Medicare’s AI prior-authorization experiment
↗ https://www.statnews.com/2026/11/12/new-documents-raise-more-questions-about-medicares-ai-prior-authorization-experiment
ONC, CMS kick off testing for electronic prior authorization
↗ https://www.fiercehealthcare.com/regulatory/onc-cms-launch-electronic-prior-authorization-testing-initiative
About the author
Amara's entry point into AI was an epidemiology role at a London research hospital, where she spent five years studying how digital health tools reached — or conspicuously failed to reach — underserved communities. Watching early algorithmic systems in healthcare quietly entrench existing inequalities, she redirected her career toward the systemic consequences of AI at scale. She covers AI through an unflinching lens: who benefits, who bears the cost, and what evidence actually says versus what the press release claims. Her writing is calm and precise, but she doesn't mistake balance for neutrality.
More from The Steward →This Week's Edition
13 November 2026
14 articles
Related Articles

Europe's Digital ID Push Tests a New Privacy Tool: Zero-Knowledge Proofs
Policy & Regulation · 5 min

Why the Woman Who Fired Sam Altman Still Doesn't Trust AI's Boardrooms
Policy & Regulation · 5 min

Pakistan to Send Climate Minister to India for First High-Level Visit in Three Years
Policy & Regulation · 5 min
Related Articles

Europe's Digital ID Push Tests a New Privacy Tool: Zero-Knowledge Proofs
Policy & Regulation · 5 min

Why the Woman Who Fired Sam Altman Still Doesn't Trust AI's Boardrooms
Policy & Regulation · 5 min

Pakistan to Send Climate Minister to India for First High-Level Visit in Three Years
Policy & Regulation · 5 min
More Stories
© 2026 Cedar & Bloom. All rights reserved.