US Medicare is testing AI to review care: What happens when algorithm decides coverage?

The programme, called the Wasteful and Inappropriate Service Reduction Model (WISeR), was launched by the Centers for Medicare & Medicaid Services (CMS) on January 1, 2026. The six-year model covers Original Medicare beneficiaries in Arizona, New Jersey, Ohio, Oklahoma, Texas and Washington.

CMS says WISeR combines artificial intelligence and machine learning with human clinical review to reduce unnecessary care, lower Medicare spending and speed up the review process.

But, what happens when an AI-assisted system influences whether an older American receives coverage for a medical procedure?

The model covers a limited number of services, including skin and tissue substitutes, electrical nerve stimulator implants and knee arthroscopy for knee osteoarthritis. CMS says these services were selected because of concerns about waste, fraud or inappropriate use.

The agency says WISeR does not change Medicare’s existing coverage rules. Emergency services, inpatient-only services and treatments where a delay could pose a substantial risk to patients are excluded.

However, records obtained by the Electronic Frontier Foundation (EFF) through a Freedom of Information Act lawsuit have raised questions about delays, technical problems, denials and the role of AI in the decision-making process.

How does Medicare’s AI experiment work?

WISeR operates through six private companies selected by CMS:

• Cohere Health – Texas
• Genzeon – New Jersey
• Humata Health – Oklahoma
• Innovaccer – Ohio
• Virtix Health – Washington
• Zyter – Arizona

The companies use technology, including AI and machine learning, to review requests for the selected services.

Under the model, providers can submit a prior-authorisation request before a service is provided or undergo a post-service, pre-payment review. The technology helps assess the request against Medicare’s existing coverage requirements.

CMS says that AI does not make the final non-payment decision. Recommendations for non-payment must come from appropriately licensed clinicians using standardised, transparent and evidence-based procedures.

The programme is not designed to allow a computer to independently deny a patient’s treatment. However, technology can still influence the process. If an AI-assisted system flags a request or recommends further review, the clinician handling the case may rely on that assessment.

That makes the accuracy of the underlying data, the system’s methodology and the quality of human oversight important parts of the experiment.

Why is CMS using AI?

CMS says the US healthcare system wastes significant amounts of money on services that may provide little or no clinical benefit in some cases. The services selected for WISeR illustrate why the agency is focusing on particular areas.

KFF found that services covered by WISeR accounted for $12.3 billion, or 5.3%, of traditional Medicare Part B spending in 2024. Skin substitutes alone accounted for $10.3 billion, or 83%, of spending on those services.

Spending on skin substitutes also rose sharply. KFF found that average spending per service increased from about $2,300 in 2019 to $21,200 in 2024, an increase of about 820%.

About 1.1 million traditional Medicare beneficiaries nationwide received at least one WISeR service in 2024, according to KFF.

CMS’s states that closer review could prevent Medicare from paying for services that are unnecessary or inappropriate, potentially reducing costs while protecting patients from treatments that may not benefit them.

KFF, however, said the financial impact of WISeR could be relatively modest initially because the selected services represent a small share of total traditional Medicare Part B spending.

The impact could become larger if CMS eventually expands the model to additional services or states.

What problems have emerged?

The biggest concern so far is delays in obtaining decisions. CMS says WISeR vendors should respond to prior-authorization requests within 72 hours.

But records obtained by EFF include cases that took considerably longer. One internal status report cited by the organisation described a request that remained unanswered for 83 days.

The records also contain complaints from healthcare providers about patients waiting weeks for approvals, with some procedures reportedly delayed or cancelled.

One provider wrote that patients were “crying in pain” while waiting for approval or guidance connected to the programme.

These complaints do not establish that every WISeR request is delayed or that AI directly caused every reported problem. But they illustrate the risks that can emerge when a new technology system becomes part of the healthcare approval process.

EFF also reported that two WISeR vendors denied 5,944 prior-authorization requests during the first three months of the programme. Its review found that Virtix denied more requests than it approved during that period.

MarketWatch separately reported high denial rates and said one vendor had been placed on a corrective action plan. 

A denial does not necessarily mean that a patient permanently loses access to treatment. Decisions can be appealed and may be reversed. But even a temporary denial can create a delay, which can be significant when treatment is time-sensitive.

Are financial incentives a concern?

The payment structure for WISeR vendors has also attracted scrutiny. CMS says participating companies receive payments linked to spending they determine to be wasteful or inappropriate and prevent from being paid. Payments also depend on performance measures, including provider experience.

EFF states that the arrangement could create a financial incentive to deny care. Its review of CMS records found that vendors receive payments for requests they deny, although the payment does not apply to denials later overturned on appeal.

CMS, meanwhile, says its payment system includes quality measures intended to balance savings with accuracy and service quality.

EFF’s review of CMS records also raised concerns about the speed of the rollout. CMS launched WISeR in January 2026, roughly six months after announcing the model.

According to EFF, Innovaccer warned CMS before launch that it had not had enough time for full end-to-end testing with providers. The company cited changing requirements, unclear governance and limited preparation time.

Innovaccer told CMS that it planned to use a system that would automatically affirm requests until it could develop full functionality. The company said “auto-affirming is the only path available” because CMS had decided not to delay the launch.

EFF’s records also describe technical and communication problems after the programme began. CMS has maintained that WISeR includes human clinical review and excludes emergency services and treatments where delays could pose substantial risks.

What does WISeR mean for Medicare patients?

For now, most Medicare beneficiaries will not be affected. WISeR applies only to Original Medicare beneficiaries in six states. It does not apply to people enrolled in Medicare Advantage.

Even in participating states, the model covers only selected medical services. For patients subject to WISeR, the main change is that certain services may require prior authorization or pre-payment review before Medicare pays.

Providers must submit information supporting medical necessity. The technology assists with the review, while a licensed clinician handles recommendations for non-payment.

If a request is denied, patients and providers can use the applicable Medicare appeals process. The key issue is how quickly the system reaches a decision.

CMS expects decisions within a defined timeframe, while some of the records reviewed by EFF describe much longer waits. For patients who need treatment, the difference between a decision within days and one that takes several weeks can be significant.

Could WISeR expand?

The current model is scheduled to run through 2031 and covers selected services in six states. CMS has indicated that the programme could provide a way to use more private-sector technology in Medicare operations.

KFF states that the model could affect more beneficiaries if CMS expands it to additional services and states. 

EFF has also pointed to planning documents that considered services such as air ambulance transport, cancer treatment, MRI scans and certain medications. Any such expansion could raise higher stakes because the consequences of an incorrect or delayed decision vary significantly by treatment.

A delay involving a non-urgent procedure is different from a delay involving cancer treatment or emergency transportation.

What happens if the AI-assisted system gets it wrong?

AI can process large amounts of information quickly, but speed does not guarantee accuracy. A medical record may not capture every factor a doctor considers important, and an automated assessment can be affected by incomplete information.

That is why CMS requires human clinical review for non-payment decisions. But human oversight is effective only if reviewers have enough information, time and authority to question an AI-assisted recommendation.

EFF has called for greater transparency around the systems used by WISeR vendors, including their testing, accuracy and potential for errors or AI “hallucinations”.

CMS maintains that the programme uses evidence-based clinical review and existing Medicare coverage policies.

Why does Medicare AI experiment matter?

WISeR could become an important test case for government use of AI in healthcare. Private insurers have used automated systems for years to review claims and prior-authorization requests, but those systems have faced scrutiny over denials and delays.

KFF found that 69% of US adults with health insurance consider prior authorization a burden, while 34% identified it as their biggest burden beyond healthcare costs when seeking care.

WISeR brings a similar process into parts of traditional Medicare, which has historically relied less on prior authorization than Medicare Advantage and many private insurance plans.

If the experiment reduces unnecessary spending without creating significant access problems, CMS could have evidence for expanding technology-assisted review.

If delays, technical problems or inappropriate denials persist, the programme could face questions about whether its savings justify the additional administrative burden and potential risks to patients.

CMS says AI-assisted review combined with human clinical oversight can reduce waste and protect Medicare funds. KFF’s analysis suggests the programme’s initial financial impact may be limited, while EFF’s review of internal records has highlighted delays, technical problems, denials and concerns about vendor incentives.

For Medicare patients, however, the question is that can they receive medically necessary care, how quickly will a decision be made, and what happens if coverage is denied?

As WISeR moves through its six-year trial, those outcomes may ultimately matter more than how sophisticated the underlying AI becomes.

The success of AI in Medicare will depend not only on whether it can identify waste, but also on whether it can do so without creating new barriers between patients and the care they need.

Disclaimer: This article is intended for informational purposes only. Financial Express Digital does not encourage or endorse any particular healthcare decision, insurance choice or course of action based on the information provided. Readers should verify the latest information from official sources and consult a qualified healthcare professional, insurance provider or relevant authority before making any healthcare or financial decisions.