AMA CMAA Framework: AI-Only CPT Codes for Medical Coders

AMA's CMAA Framework: What AI-Only CPT Codes Mean for Medical Coders

Medical coding has always assumed a physician at the center -- a clinician who sees a patient, documents the encounter, and orders the service. The current CPT system was built on that assumption. But a new framework quietly advancing inside the American Medical Association is about to challenge it.

The AMA's CPT Editorial Panel is developing a coding structure called Clinically Meaningful Algorithmic Analyses, or CMAA. The premise is straightforward but consequential: some AI systems now produce medically actionable outputs with no physician directly involved at the point of service. If those outputs affect patient care, they need a billing code -- and the CPT system doesn't yet have clean language for them. CMAA is the AMA's answer.

What CMAA Actually Describes

CMAA is designed to capture services where an algorithm -- not a physician -- processes clinically relevant data and generates a result that directly informs a clinical decision. Think of AI systems that analyze retinal scans for diabetic retinopathy without a radiologist reviewing each image, or models that screen ECGs for atrial fibrillation at scale. The algorithm runs, produces a finding, and that finding goes into the patient record. No physician work at the point of service is required.

This is meaningfully different from the AI-augmented codes introduced in CPT 2026, such as 75577 for AI-assisted coronary CT analysis (Cleerly's HeartFlow-adjacent product) or the multimodal AI evaluation codes 0992T and 0993T. Those codes describe AI as an enhancement to physician work. CMAA describes AI as the service itself.

The AMA updated Appendix S -- its taxonomy for classifying AI technology in CPT -- in early 2026 as a precursor step, establishing the classification infrastructure before the actual CMAA codes are finalized. The framework is still moving through the CPT Editorial Panel process, with applications already arriving for services that rely entirely on algorithmic output.

Why This Creates a New Coding Challenge

For coders, CMAA opens territory that existing CPT conventions don't map cleanly. Most CPT codes for diagnostic services tie reimbursement to physician work -- the professional component -- and technical execution. The relative value units are calibrated accordingly. An AI-only service has a technical component (running the algorithm, generating the report) but no traditional professional component in the physician-work sense. That gap raises several questions coders will need to navigate:

• Who bills? If no physician performed the service, the billing entity is typically the facility or the AI vendor operating under a provider agreement. Coders need to know which entity is submitting the claim.

• What documentation satisfies payer review? Without a physician note, the algorithm's output report becomes the primary documentation. Coders need to understand what those reports must contain to support the code.

• How does Medicare treat AI-only outputs? CMS has not yet confirmed coverage policy for CMAA-class services under Medicare Part B. Until it does, billing these codes risks denial under the current "AI-only" denial rationale -- where payers reject claims that show AI analysis without documented physician review.

• Which payers will follow? Commercial payers will set their own policies. Coding teams will need payer-specific coverage rules, not just CPT code knowledge.

The Precedent Already Set by FDA Clearances

There are more than 950 FDA-cleared AI/ML-enabled medical devices as of early 2026, and a meaningful share of them operate in an autonomous or near-autonomous mode -- generating findings that clinicians receive rather than perform. Reimbursement has lagged clearance significantly, partly because CPT had no framework to describe what these devices do when they run without direct physician involvement.

CMAA attempts to close that gap. If finalized and adopted into Medicare, it would mean that AI screening tools -- for diabetic retinopathy, skin lesion detection, chest X-ray triage, and similar applications -- could be billed under CPT codes that accurately describe the service without forcing the service into an existing physician-work code that doesn't fit. That's a significant shift in how autonomous AI enters the reimbursement system.

The AMA's own framing on CPT and AI-enabled services makes clear this is a deliberate effort to build durable coding language for a category that will only grow.

What Coders Should Be Doing Now

CMAA codes are not yet live in the CPT code set. But the groundwork being laid now -- Appendix S taxonomy updates, CPT Editorial Panel applications, CMS engagement -- means these codes could move faster than typical CPT additions once the framework is approved. Coders who wait until the codes are published to start learning the category will be behind.

Three concrete steps to prepare: First, get familiar with Appendix S and how the AMA classifies AI technology in CPT -- it's the vocabulary CMAA will use. Second, identify which AI tools your organization is already using that operate without direct physician involvement at the point of service. Those are the services most likely to be affected when CMAA codes arrive. Third, watch for CMS coverage determination announcements for autonomous AI-based diagnostics -- those will signal when Medicare is ready to reimburse CMAA-class claims.

The Broader Shift for Medical Coding Teams

CMAA represents something larger than a new code set. It's the CPT system formally acknowledging that AI can perform a billable medical service independently. That changes the scope of what medical coders need to know. Coding has always required understanding clinical context -- what the physician did, why, and how it connects to a diagnostic or procedural code. AI-only services add a new layer: what the algorithm did, what data it processed, and whether the output meets payer criteria for a covered service.

Coders who develop fluency in how AI systems generate clinical outputs -- not at the level of the code itself, but at the level of what the report documents -- will be better positioned to code these services accurately and defend them on audit. That's a skill set the profession hasn't needed at scale before.

Preparing Your Coding Workflow Before CMAA Arrives

The organizations that will adapt fastest to CMAA are those already building AI-aware coding workflows -- teams that understand which services in their encounter data involve algorithmic analysis and how to document those services for compliance. That kind of workflow integration, applied consistently across an encounter volume, is exactly what agentic AI coding platforms are built to support.

If your team is thinking through how to position for the next wave of CPT changes, Medikode's automated medical coding platform is built to adapt with the evolving code set -- so your revenue cycle stays current as AI-only services move from framework to live codes.