Taxonomy for Artificial Intelligence in Medical Services and Procedures provides guidance to applicants and other CPT stakeholders in determining the appropriate terminology for CPT code descriptors. It helps classify AI-enabled medical services and procedures into one of three categories: assistive, augmentative, or autonomous. This guidance should be consulted for code change applications (CCAs) which describe use of AI-enabled medical services and/or procedures.
Since the introduction of Appendix S in 2021 the use of AI in medical services has continued to evolve. At its May 2026 meeting, the CPT® Editorial Panel accepted revisions to clarify and strengthen the taxonomy. These revisions were the result of extensive feedback from Panel members and stakeholders over several months, in addition to the over four years of real-world feedback gathered from applications submitted using the existing Appendix S. The updates enhance the clarity and distinction between the assistive, augmentative, and autonomous categories, providing a more precise definition of what constitutes a clinically meaningful output.
Three categories for AI applications
This taxonomy provides guidance in determining the appropriate terminology for CPT code descriptors, including the classification of various artificial intelligence (AI) applications (e.g., expert systems, machine learning, algorithm-based services) for medical services and procedures as assistive, augmentative, or autonomous. Note that there is no single product, procedure, or service for which the term “AI” is sufficient or necessary to describe its intended clinical use or utility; therefore, the term “AI” is not defined in this taxonomy. In the CPT code set, AI software output(s) must be useful in the diagnosis, cure, mitigation, treatment, or prevention of disease or other conditions. The software’s taxonomic classification is determined based on its output(s) and the role of its output(s) in clinical care.
Assistive classification
Software is assistive when the software output(s) provides clinically relevant data without deriving a parameter (quantitative or categorical [e.g., index, score, classification]) and does not generate an interpretation or provide conclusions.
The output from assistive software requires physician or other QHP interpretation and report.
Assistive software output(s) are clinically supportive and improve physician or other QHP performance (i.e., accuracy, precision, and interobserver variability) even though the output of the primary service may be unchanged. Assistive software output(s) may include terms such as “likelihood of”, “suggestive of”, or “risk for”.
Augmentative classification
Software is augmentative when the software output(s) represent a quantitative or categorical parameter that is qualitatively different than the input. The output must be more than a summation of quantitative data inputs. The output must provide something beyond adding, averaging, measuring, or otherwise reporting descriptive statistics.
Augmentative software output(s) are reported as clinical scales, indexes, categorical classifications, risk scores, or other metrics that may be used in the diagnosis, cure, mitigation, treatment, or prevention of a disease or other condition. The output should be validated by equivalence to a metric currently in clinical use. If the output is novel, i.e., there is no such metric currently in clinical use, it should be validated for impact on patient management (e.g., novel predictive or prognostic indices). Software is classified as augmentative if the output does not include an independent interpretation or conclusion as would be required for use of autonomous.
The designation of software output as augmentative is based on demonstration that the software output is clinically meaningful and distinct from the input function.
Augmentative software output(s) are clinically meaningful based on documentation that the output from the software contributes to patient management.
Software with augmentative outputs may or may not require physician or other QHP work in the form of interaction with the software during the process between input and output. This interaction may, for example, involve physician adjustment of the settings of the software based on clinical context.
The physician work related to augmentative services typically is captured by existing codes. For example, the output parameter(s) might be a data element in an evaluation and management service, a factor in pre-surgical planning, or it may otherwise be integrated into a separate service that includes physician or other QHP interpretation.
Autonomous classification
Software is autonomous when the software automatically (without concurrent physician or other QHP involvement) derives parameters and independently generates clinically meaningful interpretations.
Reporting of the derived parameters is essential for transparency and explainability at all autonomous levels. Recommendations for definitive diagnostic conclusions or specific management or interventions should provide clinical utility and have demonstrated validity.
This can include putting the output within the context of known epidemiologic data or existing clinical practice guidelines. These standards are important for all autonomous software output(s) but most particularly for Levels II and III (as described on this page). The work of the algorithm may or may not include acquisition, preparation, and/or transmission of data. The output should establish a definitive diagnosis or recommend specific management or intervention, or subsequently also initiate management or interventions (diagnostic or therapeutic). There are three autonomous levels with varying physician or other QHP professional involvement:
Level I. The output of autonomous software includes recommendation(s) for medical management based on interpretations of, and conclusions drawn from, derived parameters. The output from the software shall allow a reasonable opportunity to negate the impending action prior to implementation (e.g., by means of an alert).
Level II. The output of autonomous software includes recommendation(s) for medical management based on interpretations of, and conclusions drawn from, derived parameters. The output from the software shall allow a reasonable opportunity to negate the impending action prior to implementation (e.g., by means of an alert).
Level III. The output of autonomous software automatically initiates management actions based on interpretations of, and conclusions drawn from, derived parameters. These medical management actions require physician or other QHP oversight and review of performance and will continue unless a physician or other QHP chooses to intervene. While Level III software implements recommendations automatically, physician oversight is warranted, typically over multiple interventions to determine whether medical management goals are being achieved.
Key definitions
In Appendix S, derived parameters are software output(s) that are quantitative or categorical (eg, index, score, classification).
In Appendix S, software output(s) are clinically meaningful based on documentation that the output from the software contributes to patient management.
In Appendix S, use of the words “automated” or “automatically” refers to the work of the algorithm in getting from input to output (ie, deriving the parameters).
Download Appendix S Table
- Appendix S table (PDF)
