Skip to content
AI in Healthcare

Article

Medicare and AI in 2026 — what CMS is (and isn't) paying for

NTAP add-on payments, Category-III CPT codes, MAC-level Local Coverage Determinations, and the MedPAC signal. What CMS is paying for AI in 2026 — and what it isn't.

By AI in Healthcare Editorial Updated
  • medicare
  • CMS
  • reimbursement
  • NTAP
  • CPT
  • LCD
  • MedPAC
  • payer
  • policy
  • clinical-ai

The single most common question a healthcare-AI vendor gets from a hospital procurement team in 2026 is not “does it work?” It is “will Medicare pay for it?” The clinical evidence base has matured faster than the reimbursement infrastructure, and CMS’s 2026 posture is best described as cautious plurality — a handful of tools are being paid for through defined mechanisms, a larger set are being tested, and most are still buried inside the DRG or the office visit.

This piece walks through the four Medicare pathways where AI is actually visible in 2026: NTAP inpatient add-on payments, Category-III CPT codes (the emerging-technology track), MAC-level Local Coverage Determinations, and the MedPAC signal — the recommendations that shape where CMS will move next. Along the way, the piece flags which pathways are working for which product categories, where the gaps are, and how a health system’s finance team should read the 2026 picture.

NTAP — the closest thing to a “Medicare paid for the AI” story

The New Technology Add-on Payment (NTAP) program is the most direct mechanism through which Medicare has paid for AI-adjacent clinical technology. NTAP was designed to give CMS a way to reimburse inpatient use of a technology that is not yet well-captured in the DRG payment weights — typically because the technology is too new for the underlying cost data. An NTAP grants a temporary, capped, per-case add-on payment on top of the DRG. It expires after roughly three years, at which point the technology’s cost is expected to be absorbed into the DRG recalibration cycle.

For AI, the two most-cited NTAP examples are stroke triage software (Viz LVO, granted NTAP in 2020) and downstream extensions. CMS publishes the current NTAP roster and payment rates in the annual IPPS final rule, which is the authoritative document for what is being paid and at what level. As of the FY2026 rule, a handful of AI-labeled or AI-heavy technologies remain on the NTAP roster; several have graduated off it (their costs are now inside the DRG); a slightly larger number are entering.

Three things to understand about NTAP as it applies to AI:

  • It is a hospital-side, inpatient mechanism. NTAP does not touch outpatient AI, imaging AI on an outpatient basis, ambient scribes, or physician-office LLM copilots.
  • It is temporary and capped. The add-on payment is a fraction of the technology’s cost, not full reimbursement, and it sunsets. Vendors and health systems that model long-term ROI against NTAP alone build a fragile business case.
  • Approval is not a coverage decision. NTAP means “we will pay extra when this is used.” It does not mean “we are asserting this should be used.” Coverage still runs through the underlying benefit categories.

For vendors of inpatient-facing AI, NTAP remains the cleanest path to a Medicare payment surface, but it is a narrow one. For CMIOs modeling AI ROI against Medicare revenue, NTAP is a temporary tailwind, not a durable one.

Category-III CPT codes — the emerging-technology track

Outside the inpatient world, Medicare payment runs through the Physician Fee Schedule and the CPT code set. Under the CPT framework, three tracks matter for AI:

  • Category-I CPT — the permanent codes that show up in the fee schedule with published Medicare payment amounts.
  • Category-III CPT — temporary codes (four-year renewable) intended for emerging technology and services. Category-III codes exist to track utilization; they can carry payment but often do not, and MACs decide coverage locally.
  • Proprietary Laboratory Analyses (PLA) codes — a specialized track for lab tests, including AI-augmented in-vitro diagnostic tests.

The AMA CPT Editorial Panel has been actively opening Category-III codes for AI-augmented services since 2021, and by 2026 the panel has a running docket of AI-specific code applications. Category-III codes have been assigned for AI-augmented interpretation of ECGs, AI-assisted mammography workflows, AI-augmented ophthalmology screening (retinal disease), and several others. Each Category-III code has its own carrier-priced reimbursement — meaning individual Medicare Administrative Contractors decide the payment.

The Category-III track is doing important structural work in 2026. It gives CMS and payers utilization data on how AI-augmented services are actually being used, on which patient populations, in which settings. That data is the input to eventual Category-I conversion, DRG rate adjustment, and coverage-policy decisions. Without it, the reimbursement conversation would be based on vendor claims alone.

But two frictions are real:

  • Category-III codes often pay poorly or not at all. A code that exists but pays $0 does not create a business case; it creates a data-collection burden.
  • Category-III-to-Category-I conversion is slow. Codes can sit in Category-III for years before the panel is comfortable moving them. That is by design — the framework exists precisely to hold new services in a probationary track — but it means the reimbursement story for AI is a multi-year one.

MAC-level Local Coverage Determinations — where the real fights happen

Medicare’s coverage architecture is federated. The Medicare Administrative Contractors — regional entities that process claims and make coverage decisions for their geographic areas — issue Local Coverage Determinations (LCDs) that decide, at the regional level, which services are considered reasonable and necessary. When there is no National Coverage Determination for a service, the LCDs are the decision surface.

For AI, LCDs are where a lot of the operational reimbursement fights are actually happening in 2026. The CMS Medicare Coverage Database publishes active and proposed LCDs, and AI-related LCDs have been steadily accumulating — for AI-augmented cardiac imaging, AI-based skin lesion analysis, AI-augmented retinal disease screening, AI in colonoscopy CADe, and others.

Two dynamics matter:

  • Regional variation is real. A service can be covered by one MAC and non-covered by another. Vendors of MAC-priced AI services face an actual patchwork of regional coverage decisions.
  • The evidence bar is set locally. MAC medical directors have discretion over what evidence they find persuasive. Peer-reviewed multi-site RCTs are the strongest currency; single-vendor pivotal studies are a weaker one. Vendors whose evidence base is thinnest often see the widest MAC-to-MAC variation.

For a national health-system operator with facilities in multiple MAC regions, the practical implication is that “does Medicare cover this?” is not a single-answer question. Revenue-cycle teams need MAC-level coverage tables, updated on the LCD cadence.

The MedPAC signal — where CMS is likely to move next

The Medicare Payment Advisory Commission (MedPAC) is the independent congressional body that recommends Medicare payment policy. Its recommendations are not binding, but historically they are a leading indicator of CMS direction. MedPAC’s June and March reports are published in full and are the durable reference for reading the direction of travel.

Three MedPAC threads to watch through 2026 and into 2027:

  • AI in Medicare Advantage risk adjustment. MedPAC has consistently expressed concern about coding intensity in Medicare Advantage, and AI-assisted coding tools are now sufficiently prevalent that they show up in the actuarial commentary. Watch for MedPAC recommendations that push CMS toward risk-model adjustments (see the PwC 2027 medical-cost-trend commentary for the parallel commercial dynamic).
  • AI-augmented prior authorization. MedPAC has flagged prior-authorization automation on the payer side as an area needing utilization guardrails. Watch for recommendations that constrain how AI-based coverage-decision tools are used in Medicare Advantage and traditional Medicare adjacencies. This is the payer-AI topic hub in its Medicare form.
  • Ambient documentation and E/M level distribution. As ambient-scribe deployment scales inside physician offices, MedPAC is beginning to look at whether the E/M level distribution is shifting in ways that reflect the tool rather than the visit. This is early — the data is still thin — but it is the same coding-density story that the commercial actuarial community has already surfaced.

Practical implications for provider organizations

For hospital and health-system finance and IT leaders, the 2026 CMS picture translates into four operational imperatives:

  1. Do not model AI ROI against a single reimbursement pathway. NTAP is temporary. Category-III codes often do not pay. MAC coverage is regional. A resilient AI business case rests on cost-of-care effects, revenue-neutrality proofs, and — increasingly — on productivity effects, rather than on a single line item of Medicare add-on payment.
  2. Track LCDs by MAC. Revenue-cycle teams should have an LCD table for AI-augmented services, updated on the LCD cadence, and know which of their facilities sit in which MAC region.
  3. Watch for the MedPAC direction of travel on risk adjustment. Health systems with heavy Medicare Advantage exposure and heavy AI-documentation deployment should model both directions — the current coding-density tailwind and the medium-term recalibration risk.
  4. Coordinate with your governance program. The Medicare reimbursement question sits alongside the FDA regulatory question, the Joint Commission RUAIH reporting question, and the payer-contract question. These conversations should not be in separate rooms.

What is not on the CMS reimbursement map in 2026

For completeness, several categories where AI is being deployed at scale but where Medicare reimbursement is either absent or invisible:

  • Ambient scribes. No dedicated payment surface; effect flows through visit-level E/M coding.
  • In-basket message drafting and inbox triage. No dedicated payment surface.
  • Chart summarization / Note Buddy-style features. No dedicated payment surface.
  • General LLM copilots for clinicians. No dedicated payment surface.
  • Most patient-facing chat and symptom-triage tools. Outside the Medicare payment envelope entirely.

That absence is not a permanent state — but it is the state in 2026, and any business case pinned on near-term CMS reimbursement for these categories is on shaky ground.

Sources cited