When Algorithms Deny Care

This month, roughly 1,000 pages of internal federal records were released following a Freedom of Information Act (FOIA) lawsuit by the Electronic Frontier Foundation (EFF). The documents pulled back the curtain on one of the federal government's most ambitious experiments in healthcare technology: Medicare’s Wasteful and Inappropriate Service Reduction (WISeR) pilot.

What was promised as a cutting-edge effort to reduce administrative waste and root out fraud quickly became a cautionary tale of care denials, multi-week backlogs, and patient suffering.

Here is an accessible look at what happened, why it went wrong, and what it means for the future of AI in American medicine.

What Was the WISeR Pilot Supposed to Do?

Historically, traditional fee-for-service Medicare (covering Parts A and B) rarely required "prior authorization" (the process where a doctor must ask an insurer for permission before performing a procedure). Prior authorization was largely confined to private Medicare Advantage plans.

That changed on January 1, 2026, when the Centers for Medicare & Medicaid Services (CMS) launched WISeR in six states: Arizona, New Jersey, Ohio, Oklahoma, Texas, and Washington. Affecting roughly 6.4 million seniors and people with disabilities, the six-year program required doctors to seek prospective approval for 13 to 16 specific treatments, including:

• Epidural steroid injections for chronic pain.

• Advanced biologic skin substitutes for diabetic wounds.

• Minimally invasive knee arthroscopy and spinal decompression surgeries.

• Implantable nerve stimulators for sleep apnea and incontinence.

• Kyphoplasty and vertebral augmentation to stabilize painful spinal fractures.

Instead of using regional Medicare claims staff, CMS hired private health-tech vendors to evaluate these requests using artificial intelligence and machine learning algorithms. The goal was rapid, automated approvals for routine care and quick detection of inappropriate claims.

What Actually Happened: The Reality Behind the Rollout

Instead of streamlining care, the pilot created immediate administrative friction. The newly unredacted federal records highlight four major points of failure:

1. A Surge in Rejected Care

During the first three months of 2026, technology vendors denied 5,944 requests for care. In Washington state, contractor Virtix Health rejected 53% of all prior authorization requests across more than 6,000 cases. It was denying more treatments than it approved, far exceeding CMS’s initial projection of a 10% to 22.5% reduction.

2. Endless Wait Times

CMS told the public that requests would be resolved within 72 hours (or 48 hours for urgent cases). In practice, requests regularly stalled for weeks. One documented authorization request sat in an unreviewed queue for 83 days.

3. Unready Software Rushed to Market

WISeR was announced in June 2025 and launched just six months later. In December 2025—weeks before the start date—vendor Innovaccer warned CMS that its systems were not fully functional and had not undergone end-to-end testing with healthcare providers. When CMS refused to delay the rollout, the vendor had to resort to an emergency workaround in Ohio: automatically approving ("auto-affirming") incoming claims to keep the system from crashing entirely.

4. Financial Incentives That Rewarded Denials

Perhaps the most controversial revelation involves how vendors were paid. CMS structured contracts so that vendors received contingency fees of up to 25% of the baseline cost of denied procedures. While CMS docked pay for low customer service or delay scores, those penalties amounted to only 5% to 10%. Vendors had a direct financial incentive to reject care.

Key Pilot Metric	Contractor Outcome	Official Program Target
Washington State Denial Rate	53% initial denials (Virtix Health)	10% to 22.5% projected reduction
Total Q1 Vendor Denials	5,944 care requests denied	Baseline clinical exceptions only
Adjudication Speed	Backlogs lasting weeks; one request took 83 days	Mandatory response within 72 hours
Vendor Contingency Pay	Up to 25% of denied procedure costs	Neutral administrative fee
Quality Performance Penalty	Deductions of only 5% to 10% for poor service	Strong deterrent against improper denials

The Human Toll: Patients Waiting in Pain

Behind the statistics are elderly and disabled patients stranded without medical relief. Provider complaint logs submitted to CMS describe clinical gridlock:

• Spine and Fracture Care: Patients with osteoporotic spinal fractures require prompt kyphoplasty to stabilize bone fragments and relieve severe pain. Instead, authorizations stalled. One doctor wrote to CMS: "I have had to watch 3 patients cry at bedside for not hearing back on their prior auth for kyphoplasty/vertebral augmentation procedure. These patients are in deep pain."

• Wound Care: Complex diabetic foot ulcers require timely cellular skin substitute grafts to prevent bone infection and potential amputation. Administrative delays forced surgeons to repeatedly postpone procedures or risk performing uncompensated care.

The Pushback: Courts, Congress, and States React

The September 2026 disclosures ignited an immediate regulatory backlash:

• GAO Rule Classification: The Government Accountability Office (GAO) determined that CMS improperly bypassed Congress by creating new coverage rules without required notice-and-comment procedures.

• Congressional Action: In the House Ways and Means Committee, lawmakers forced an amendment requiring CMS to publicly release all internal vendor audits and Corrective Action Plans within three months. Meanwhile, an appropriations rider introduced in the House seeks to block all ongoing funding for the WISeR program.

• State Laws Banning AI-Only Denials: States across the country are stepping in to protect patients. Washington (SB 5395), Indiana (HB 1271), Iowa (HF 2635), Alabama (SB 63), and California (SB 1120) have passed laws mandating that an algorithm can never be the final decision-maker on medical necessity; only an actively licensed physician can issue a denial.

Key Takeaways for the Future of Healthcare AI

The failure of the WISeR pilot does not mean technology has no place in healthcare operations. Instead, it exposes what happens when automation is deployed without clinical safeguards.

1. Remove the Profit Motive from Denials: Paying vendors a bounty for denied care turns utilization review into automated rationing. Payer-vendor contracts must use neutral, flat-fee models that reward accuracy and speed, not rejections.

2. Keep Humans in the Loop: Machine learning can help organize records and flag missing information, but the final determination of medical necessity must always rest with a qualified physician in the appropriate specialty.

3. Integrate Directly into Clinical Workflows: Forcing medical practices to manually upload charts to third-party web portals creates administrative chaos. Modern systems must use open interoperability standards (such as HL7 FHIR) to communicate directly with electronic health records.

4. Test Thoroughly Before Going Live: Healthcare software cannot follow the Silicon Valley mantra of "move fast and break things". When systems break in clinical settings, vulnerable patients pay the price.

The lesson of Medicare's WISeR experiment is clear: automated efficiency must never come at the expense of patient care and transparent clinical oversight.