The AMA released CPT 2026 with 288 new codes, and for the first time in the 50-year history of procedural terminology, artificial intelligence has its own billing language. That is genuinely significant. What is not significant: most of it doesn’t pay.
CMS assigned status indicator “I” — not valid for Medicare purposes — to the majority of new AI-related CPT codes effective January 1, 2026. The few that achieved Category I status with real pricing are a narrow set of imaging-based AI analyses: coronary plaque assessment, perivascular fat analysis for cardiac risk, and burn wound multispectral imaging. Everything else — the clinical decision support, the AI-augmented documentation, the algorithm-driven risk stratification that 75% of U.S. health systems are now deploying — sits in Category III. Temporary codes. No national Medicare rate. No guaranteed coverage.
Health systems are purchasing AI platforms faster than they can govern them. Now they are learning that “code it” and “bill for it” are two different statements.
Clinical Context: The Gap Between Code and Payment
The CPT code set is a language, not a payment guarantee. Category I codes carry established Medicare valuation. Category III codes exist to track utilization of emerging technologies — they create a data trail so that CMS can eventually value them. The gap between those two states is typically three to five years, sometimes longer.
CMS made the gap explicit this cycle. When the AMA submitted updated practice expense survey data for 2026 — data that, for the first time, included AI licensing costs, SaaS subscriptions, and algorithm-based service expenses — CMS rejected it. The agency cited “quality concerns” with the updated methodology and confirmed it would continue using 2007–2008 data as the basis for practice expense calculations. That means CMS is pricing the clinical labor of AI-assisted medicine using cost data from a world before the iPhone existed.
The AMA is already working on a fix: a proposed new classification called Clinically Meaningful Algorithmic Analyses (CMAA) codes, specifically designed to capture AI-SaaS product costs within the CPT framework. The CMAA initiative is still in development. The codes are not live. The clinical AI market is.
This is the compliance and financial trap sitting in front of every health system CFO right now: AI expenses are real, escalating, and not recoverable under current CMS policy.
Four Findings Healthcare Leaders Must Absorb
First-ever AI CPT taxonomy — still without national pricing for most services. CPT 2026 introduced Appendix S, a formal taxonomy classifying AI services as assistive (detects patterns, clinician decides), augmentative (quantifies data, clinician interprets), or autonomous (acts without real-time clinician input). This taxonomy is clinically useful. It is not accompanied by CMS fee schedule pricing for the majority of codes that use it. Of the new AI-related additions, only the coronary atherosclerotic plaque assessment codes crossed from Category III to Category I with finalized Medicare payment. Everything else awaits future rulemaking.
CMS explicitly declined to modernize its practice expense methodology for AI. The CY 2026 Medicare Physician Fee Schedule Final Rule confirmed CMS would not adopt updated AMA survey data capturing SaaS and AI tool costs, citing data quality concerns. The current practice expense framework — built on inputs from 2007 and 2008 — cannot account for subscription-based AI tools, algorithm licensing fees, or continuous model monitoring costs. CMS acknowledged the gap and indicated it “may propose changes in future rulemaking.” That is not a timeline.
Shadow AI is compounding the governance problem at scale. 75% of U.S. health systems are using or plan to deploy AI platforms, but only 10% use automated monitoring to detect which AI capabilities are actually running across their enterprise. A Wolters Kluwer survey of 500+ healthcare professionals found 40% had encountered unauthorized AI tools at work and nearly 20% admitted to using them. One audited mid-size system with 8 hospitals discovered 23% of clinicians were regularly using consumer AI tools for clinical documentation — outside any HIPAA agreement, outside any sanctioned workflow, outside any billing or governance framework. Healthcare data breaches now average $7.4M per incident. The liability math is direct.
Pharmacy AI workflows sit directly in the unpriced gap. Clinical decision support software — the category most relevant to pharmacy operations, prior authorization routing, formulary management, and medication therapy management documentation — falls almost entirely into the category of tools CMS declined to price for 2026. Pharmacy directors deploying AI for drug utilization review, discharge medication reconciliation, or prior auth automation face the same structural issue: the workflow exists, the technology exists, but the reimbursement infrastructure does not exist yet. Organizations building cost justification models for pharmacy AI solely on expected billing recovery are projecting against a payment framework CMS has not yet built.
Operational Impact: Three Immediate Obligations
Billing integrity. Do not submit Category III AI codes expecting Medicare reimbursement at a national rate. They do not have one. Any organization submitting claims for Category III AI services under the assumption of standard Medicare coverage is operating on a false premise. Revenue cycle teams need to audit current and planned AI billing against the CMS status indicator list before the end of Q2. A CPT code existing for a service does not mean a payer will cover it, does not mean a state Medicaid program will recognize it, and does not mean your existing contract rates include it. These are three separate conversations.
Cost accounting. AI platform costs — licensing, SaaS subscriptions, algorithm maintenance, integration fees — are not currently recoverable under CMS practice expense methodology. That means every dollar spent on enterprise AI is a cost center, not a recoverable expense, until CMS updates its framework. CFOs and pharmacy directors building multi-year AI business cases need to model for this correctly. The organizations that will be in the worst position are those that built ROI projections around billing recovery that does not exist and are 18 months into deployment before they figure that out.
Governance and compliance. The shadow AI data is not a technology problem — it is a policy gap. If 20% of clinical staff are using unauthorized AI tools, the organization does not have an AI strategy; it has an AI exposure. The difference matters when a breach occurs or when a state licensing board comes looking. A formal AI governance policy with an approved tool list, a shadow AI detection mechanism, and a defined review cycle is the minimum defensible position. AI enforcement models across 21 states now include fines up to $15,000 per day for non-compliant AI deployments. Informal governance is not a defense.
TheraIntel Perspective: Rewrite the Business Case
The CPT 2026 AI codes are not a breakthrough. They are a marker. They tell you where the system wants to go. CMS declining to price those codes in the same cycle tells you how far it still has to travel.
Health systems that treat this as an administrative footnote will encounter it as a financial problem 18 months from now, when the AI budget line has grown and the reimbursement pathway still does not exist. The organizations that get ahead of this are not the ones waiting for CMS to catch up. They are building internal cost justification models that do not depend on billing recovery: productivity recapture, documentation error reduction, FTE redeployment, and risk-adjusted outcome metrics.
One point that deserves direct language: most health systems selling AI deployment internally to their CFO are using billing recovery as part of the justification. That argument is, at present, mostly fictional for any AI tool outside of three narrow imaging categories. If your AI business case includes a billing revenue line that does not correspond to a Category I code with an actual CMS rate, rewrite it now.
The pharmacy workflow is directly in the gap. Prior auth intake, discharge medication reconciliation, formulary exception routing — these are the workflows being automated at scale right now, and they are exactly the workflows where CMS has not established a billing framework. Pharmacy directors who have deployed AI in these areas and have not audited their billing posture need to do that audit. Not next quarter. Now.
Three Actions Before End of Q2
Audit your AI billing posture against the 2026 CPT status indicator list. Pull every AI-related CPT code your organization is submitting or planning to submit. Cross-reference against CMS’s status indicator assignments. Any Category III code being submitted with an expectation of Medicare payment at a national rate is a compliance exposure. Fix it before the end of Q2, not after a claim denial triggers the review.
Rebuild your AI ROI model without billing recovery. The right cost justification for clinical AI in 2026 runs through operational value: time per encounter, prior auth cycle time, discharge reconciliation error rate, FTE hours recaptured. These are measurable, defensible, and do not depend on CMS updating its payment framework. Build your business case on the value you can document today.
Implement a formal shadow AI governance policy this quarter. An approved tool list, a mechanism for detecting unauthorized AI usage across clinical workflows, and a defined quarterly review cycle. This is the minimum defensible position in a regulatory environment where 21 states have active AI enforcement frameworks and healthcare data breaches average $7.4M per incident.
“CPT 2026 gave healthcare AI a vocabulary. CMS chose not to fund most of it yet. Organizations that modeled their AI ROI around billing recovery will feel that gap before CMS closes it.”