AMA Updates AI Taxonomy to Clarify Clinical Coding Standards

AMA refines framework for coding AI-enabled medical services

The American Medical Association's CPT Editorial Panel has approved significant revisions to its AI taxonomy framework, clarifying how clinical artificial intelligence applications should be classified and coded within the Current Procedural Terminology system.

The updated CPT Appendix S strengthens the foundation established in 2022 for categorizing AI medical services as assistive, augmentative, or autonomous based on how software output functions in clinical settings. The changes address confusion that emerged as stakeholders submitted code applications over the past four years.

"The original taxonomy gave us a foundation, but AI in clinical practice has moved quickly," said Dr. Richard Frank, a CPT Editorial Panel member and co-chair of the Digital Medicine Coding Committee. "The revisions sharpen the distinctions between categories in ways that reflect how the output from software is actually used by physicians to the benefit of patients today."

What changed in the taxonomy

The revisions emerged from months of consultation with industry stakeholders, payers, the Centers for Medicare & Medicaid Services, the Food and Drug Administration, medical specialty societies, and panel members who reviewed more than four years of code applications.

Key updates include refined definitions that draw clearer boundaries between assistive and augmentative services, a formal table defining terms like "derived parameters" and "clinically meaningful," and replacement of the term "machine" with "software output" throughout the document to better capture the range of AI architectures in clinical use.

The changes specifically address feedback indicating that applicants needed greater clarity about category boundaries and that physicians required better guidance on when software output could stand independently versus when physician interpretation was required.

Growth reflects clinical adoption pace

When the taxonomy was introduced to the panel in 2021, only one AI-classified CPT code existed. With the framework's guidance, 43 taxonomically-classified codes have been accepted across code years 2021 through 2028, spanning cardiology, oncology, ophthalmology, insulin dosing, and other medical fields.

"One of the most important things these revisions do is clarify what a clinically meaningful output looks like," said Dr. Mark Synovec, co-chair of the Digital Medicine Coding Committee. "That matters not just for applicants drafting code change applications, but for the integrity of the codes themselves."

Why it matters

Precise taxonomy language serves multiple critical functions in healthcare's AI adoption. Physicians need clear guidance on what level of clinical oversight algorithms require in practice. Payers must align coverage and payment policy with AI's actual clinical role. Regulators depend on stable terminology to establish durable coverage and quality policies.

The framework's technology-agnostic, output-based approach allows it to remain relevant as AI capabilities evolve, avoiding the need for continuous structural revision as new architectures emerge. Well-defined categories produce better code descriptors, which enable more accurate reporting of services delivered to patients and support appropriate reimbursement pathways.

Most AI-enabled services enter the CPT code set as Category III codes—temporary codes for emerging technologies that allow data collection on clinical efficacy and outcomes before meeting the evidence standards required for permanent Category I codes.

The American Medical Association first reported these details about the CPT Editorial Panel's actions.