CPT 2026 AI Codes: What Every Medical Coder Must Know

CPT 2026 AI Codes: What Every Medical Coder Must Know

For years, AI-assisted clinical services lived in CPT's Category III section — temporary codes, often uncovered, used mainly to collect data while payers decided whether the technology was real. That changed on January 1, 2026. The AMA promoted the first wave of AI-augmented diagnostic services to permanent Category I status, and CMS proposed Medicare payment for at least one of them. For medical coders, this is not an incremental update. It is a structural shift in how AI-assisted services get billed, documented, and audited.

If your team hasn't updated its coding protocols to account for these new codes, you are either leaving revenue on the table or running compliance risk — sometimes both.

Why 2026 Is Different for AI in CPT

Category III codes have always been a signal, not a destination. They mark emerging technologies as real enough to track, but uncertain enough to hold at arm's length from mainstream reimbursement. When a service graduates to Category I, it crosses a threshold: the AMA's Editorial Committee has determined the procedure is performed by many practitioners, the clinical data supporting it is established, and it is consistent with contemporary medical practice.

The 2026 CPT code set, released by the AMA on September 11, 2025 and effective January 1, 2026, includes 288 new codes, 84 deletions, and 46 revisions — 418 changes in total. A meaningful cluster of those changes specifically concern augmentative and assistive AI services: software that analyzes clinical data sets and delivers results that a physician then interprets and reports. These are not AI tools that replace the physician. They are AI tools that work alongside the physician, and now CPT has a permanent home for them.

The New AI-Augmented Codes You Need to Know Now

Coronary Plaque Assessment Gets a Permanent Home: CPT 75577

The most significant upgrade is the promotion of coronary atherosclerotic plaque assessment from four temporary Category III codes (0623T–0626T, in use since 2021) to a single permanent Category I code: 75577. The full descriptor reads: Quantification and characterization of coronary atherosclerotic plaque to assess severity of coronary disease, derived from augmentative software analysis of the data set from a coronary computed tomographic angiography, with interpretation and report by a physician or other qualified health care professional.

Cleerly, the Denver-based company whose AI coronary imaging platform drove much of the Category III utilization data, is the primary beneficiary of this upgrade. The company's software analyzes CCTA data sets to quantify plaque type and burden, producing reports that cardiologists use to stratify risk and guide treatment. CMS proposed Medicare payment for CPT 75577 in the 2026 Physician Fee Schedule proposal, which is a significant signal that national coverage — not just commercial insurer coverage — is moving in this direction.

For coders, the operative phrase in the descriptor is "with interpretation and report." The physician must document a separate, signed interpretation. Attaching the AI software's output to the chart without a physician narrative does not support billing 75577. This is the documentation gap most likely to trigger a denial or audit finding.

Burn Wound Imaging and Cardiac Risk: Three More to Watch

Three additional AI-augmented codes warrant attention:

• 0972T — Multi-spectral imaging for burn wounds using an algorithmic device to evaluate healing status. Available since July 1, 2025, this code officially enters the CPT manual in 2026. It covers all services related to device use, including setup, image transmission, and report generation. Coders should verify that documentation captures the algorithmic classification result and the clinical decision it supported.

• 0992T — Noninvasive cardiac risk assessment from augmentative software analysis of perivascular fat, without concurrent CT scan of the heart. Both codes require documentation of the software analysis output and the interpreting physician's report.

• 0993T — Same service as 0992T but with concurrent CT scan. The key distinction for coders is whether the CT scan was performed at the same session; getting this wrong means either unbundling or missing the add-on appropriately.

What Triggers These Codes — and What Doesn't

All four codes share a common triggering requirement: an AI or algorithmic software system must perform the analysis, and a qualified physician or other qualified health care professional (QHP) must interpret and report the results. The analysis alone does not generate a billable event. The interpretation does.

This matters because many practices are still treating AI-generated reports as automated outputs that flow into the chart without physician engagement. If the radiologist or cardiologist reviews the AI output and incorporates it into their own interpretation — and documents that explicitly — the code is supportable. If the AI output simply exists in the chart as an attachment with no physician commentary, it is not.

CPT's language of "augmentative" versus "assistive" AI is also worth understanding. Augmentative software performs the analysis; the physician adds judgment. Assistive software surfaces information to aid physician decision-making but the physician performs the core analysis. The 2026 AI codes are augmentative — the software generates the clinical output, and the physician validates and reports it. Coders who conflate these categories risk selecting the wrong code family entirely.

Documentation Requirements Are Not Optional

For every AI-augmented CPT code that went live in January 2026, documentation must support three things: the input (the data set analyzed — CCTA images, spectral imaging, CT scan), the process (which AI or algorithmic system was used, by name or description), and the output (the physician's interpretation and report, separate from the software printout).

Payers are already scrutinizing AI-assisted claims more carefully than legacy codes. The PHTI's April 2026 report on administrative AI found that health plans are increasingly deploying automated systems to detect billing intensity shifts — and AI-augmented diagnostic codes represent exactly the kind of newly minted, higher-value claim that algorithmic payer systems will flag for review. Getting the documentation right at submission is not defensive practice; it is the only way to avoid the denial and appeal cycle that consumes revenue cycle resources.

Where Payer Coverage Stands Today

CPT 75577 is the strongest candidate for near-term national coverage. CMS's proposed payment in the 2026 Physician Fee Schedule signals intent, and once the final rule is published, Medicare coverage status will clarify. Commercial payers vary: a handful of major plans already covered the predecessor Category III codes (0623T–0626T) in specific cardiometabolic risk contexts, and they are reviewing whether to extend or replace that coverage with the new Category I code.

The burn wound (0972T) and cardiac risk perivascular fat codes (0992T/0993T) remain in Category III, which means payer coverage is not guaranteed. Always verify coverage before billing. Some payers will issue a denial with a specific reason code indicating the service is investigational; others will simply deny as non-covered. Document your verification in the patient record so that any appeal is supported from the start.

What Agentic AI Can Do That Manual Coding Can't

Tracking Category III-to-Category I upgrades, verifying payer coverage changes by code, and flagging documentation gaps for AI-augmented services are exactly the kinds of pattern-matching tasks that human coders cannot sustain at scale. A coder reviewing 200 charts per day will not catch every missed 75577 opportunity or every 0992T/0993T distinction — especially when the triggering service is embedded in a complex cardiology note rather than a standalone procedure report.

Agentic AI coding platforms can be configured to recognize these triggers, surface them for coder review, and flag documentation that is insufficient to support billing. That is not replacing the coder's judgment; it is giving the coder's judgment the right information at the right moment. The AMA's 2026 CPT code release marks a turning point in how AI services are reimbursed — and the practices and health systems that adapt their coding workflows now will capture revenue that others will miss for months.

If your team is still building those workflows manually, Medikode's automated medical coding platform is built to handle exactly this kind of complexity — surfacing the right code for the right clinical documentation, across every specialty, every day.