Comment on CMS-2026-2377-0002

The infrastructure for AI reimbursement is being built right now, in real time, and it will shape incentives for a generation. Right now, AI-related reimbursement flows to the billing entity — typically a physician group, hospital, or health system — with no requirement to show how the payment is used. That's standard for CPT billing generally, and no one is proposing CMS audit every line item on every claim. But AI is not a typical add-on service. It is, structurally, different from most things CPT has historically paid for: The clinical labor is front-loaded into the tool, not the encounter. A radiologist reviewing an AI-flagged scan is often doing less independent analytic work than before, not more — the value has partially shifted from the clinician's time to the vendor's software. Vendors, not clinicians, are the primary economic beneficiary of scale. A hospital system report on hundreds of thousands of studies pays licensing fees on volume; the marginal reimbursement dollar for an AI-assisted read has an obvious first claimant sitting outside the hospital altogether. The workforce absorbing the operational load — nurses, techs, and support staff acting on AI-generated alerts, monitoring outputs, documenting AI's role per the new requirements — is not the party billing for it. New AI codes require documentation of "the specific AI tool used, its clinical purpose, and how the output was applied," which is disproportionately clinical-staff work even when the physician bills. Trade groups anticipating the 2026 codes are already framing this in almost entirely commercial terms — coding leads, RCM partnerships, "capturing" reimbursement, avoiding "revenue left on the table.The gap between clinical use and billing capture is where revenue is being lost, and the 2026 CPT updates bridge that gap for practices ready to act. That is a legitimate business concern for practices. It is also a warning sign for policy: an entire industry is already organizing around AI reimbursement as a revenue capture opportunity, with essentially no public conversation yet about whether that revenue should be required to reach the workforce delivering the care downstream of the algorithm. CMS doesn't set list prices in a vacuum — it sets the reimbursement architecture that every hospital's internal budget decisions respond to. If AI-related reimbursement is treated as ordinary margin, hospitals will do what hospitals do with ordinary margin: some will reinvest in staffing, many will use it to offset vendor licensing costs, and some will simply let it accrue as system-level revenue with no obligation to trace it back to the clinical workforce whose labor and oversight make the AI service billable at all. That outcome would be a policy failure hiding inside a coding update. AI-assisted care is being sold to CMS, to Congress, and to the public on the promise that it will free up clinical capacity, reduce burnout, and let clinicians spend more time on judgment and less on rote analysis. If the reimbursement built to reward that promise never touches nurse staffing ratios, clinical education, or frontline compensation, the promise becomes marketing, not policy. Facilities billing AI-related CPT codes above a defined volume threshold should report, at minimum, an annual breakdown of how associated reimbursement was allocated — vendor licensing and technology costs, physician/QHP compensation, and clinical workforce investment (staffing, training, wage adjustments). This does not need to be claim-level; a facility-level annual disclosure, similar to existing cost report mechanisms, would be sufficient to create visibility CMS currently lacks entirely. CMS should study — and pilot, potentially through the CMS Innovation Center — whether a defined percentage of incremental AI-related reimbursement can be conditioned on demonstrable clinical workforce investment, analogous to medical loss ratio requirements that already tie insurer reimbursement to actual patient-care spending rather than administrative overhead or profit. Even short of new rulemaking, CMS should publicly acknowledge that AI-related CPT reimbursement, as currently structured, has no mechanism ensuring gains reach the clinical workforce rather than technology vendors — and commit to studying it before the payment architecture hardens further. The 2026 code set is a foundation; each subsequent year of expansion will be harder to walk back or condition retroactively. The stakes of waiting Before the payment architecture for AI-assisted care fully hardens, require transparency about where the money goes, and treat clinical workforce reinvestment as a condition of that reimbursement.