Medicare takes a first step toward paying for the AI in your doctor's office

Medicare pays easily for a cotton swab, and it pays for the wear on a CT scanner, but it has no good way to pay for the algorithm that reads the scan. The CY 2027 hospital outpatient proposed rule, published July 2, is Medicare's first serious attempt to fix that. The mechanism is not a price. It is a label.

The proposed rule, CMS-1850-P, retires the casual "SaaS" shorthand and coins "Software as a Medical Service" (SaMS) to describe software-based technologies that produce clinical information: algorithms that flag a suspected pulmonary embolism, score coronary risk from a CT, or map prostate cancer on an MRI. CMS is saying these outputs should be paid as part of a medical service, not billed like a software license or bundled invisibly into the underlying scan.

In Medicare's hospital outpatient system, what an item is called determines whether it gets its own payment line, gets rolled into a larger bundle, or is treated as a supply the hospital eats. The same algorithm can be a billable service, a packaged cost, or a non-entity depending on which bucket it lands in. The proposed rule's SaMS framing is therefore the policy itself: once an algorithmic output is recognized as a service, the payment conversation can begin. Until it is recognized, every vendor and hospital has to argue from a different starting line.

The rule does more than rename things. CMS also proposes a 2.4% outpatient rate update for 2027 (3.2% market basket minus a 0.8 percentage point productivity adjustment) for hospitals and ambulatory surgery centers that meet quality reporting requirements, affecting roughly 3,500 hospitals and 6,400 ASCs. It would apply physician-fee-schedule rates to imaging-without-contrast services at excepted off-campus provider-based departments, which Becker's Payer Issues estimates at about $260 million in first-year Part B savings (roughly $190 million in program spending and $70 million in beneficiary premiums), with rural Sole Community Hospitals exempted.

The largest dollar move is not AI-related. Based on a 2026 acquisition cost survey, CMS proposes paying for drugs acquired under the 340B drug pricing program at ASP minus 33.4%, estimated to reduce Original Medicare drug payments by $4.55 billion and beneficiary drug payments by $1.15 billion in the first year, paid for by raising non-drug outpatient payments on a budget-neutral basis. A separate codified 0.5% reduction to the outpatient conversion factor continues to claw back the $7.8 billion in overpayments from 2018 to 2022, with CMS now signaling a shorter recoupment window.

The harder question the rule does not answer is the one the Bipartisan Policy Center has spent two years trying to frame: once an algorithmic output is recognized as a service, how is its value measured? CMMI's ACCESS model, which ties recurring payments to outcome measures for technology-supported chronic care, is the live laboratory. The OPPS proposal is not. CMS has left the valuation question to a future rule and signaled, through last summer's software RFIs and the December 2025 HHS request for information on accelerating AI in clinical care, that this is an interim step rather than a final methodology.

The rule makes algorithmic work visible to the payment system, which means clearer coding, cleaner billing, and a defined lane for vendors to argue from. It does not yet say whether a given AI tool will be paid for its sticker cost or for the patient outcomes it produces. That choice will determine whether the next wave of clinical AI gets adopted because it improves care, or because it adds a billable line.

The standard 60-day comment window opens with Federal Register publication. The final rule typically lands in early November.