Something remarkable is happening in American medicine. The FDA has now authorized roughly 1,450 AI-enabled medical devices.1 It’s one of the fastest waves of clinical innovation in a generation, bringing novel tools that can flag a stroke on a scan, catch an arrhythmia a human eye might miss, or estimate a patient’s cardiac filling pressure without an invasive catheter procedure, each catching problems earlier than would otherwise be possible.
But innovation only helps patients if it reaches them, and for a long time, the path from a brilliant algorithm to a bedside has run through a payment system that simply wasn’t built for it. A 2025 study in the Journal of the American College of Radiology tracked every Medicare Part B claim radiologists submitted for AI services from 2018 to 2023. Radiology has been at the forefront of AI adoption, and as the most mature corner of clinical AI, it’s the best test of whether the system works at all. The authors identified 23–what they dubbed “Software-as-a-Service” (SaaS)-based–AI services, of which radiologists submitted 83,392 claims against those codes over that six-year span. Medicare denied 52.6% of them, roughly $16.4 million in services rendered and never reimbursed. By 2023, as volume surged, the denial rate climbed to 68.2%.2
“…increasing AI adoption in Medicare provides empiric evidence for the need to design payment systems suitable for such services.” – Zhang et al. (2025)
None of this reflects systemic fraud or sloppy billing. It reflects a payment system built decades ago to price staff time, equipment, and supplies, not a trained algorithm whose real value comes from years of development and validation that never show up in a five-minute patient visit.
This month, CMS took a genuinely exciting step toward closing that gap.
What CMS Proposed
In the CY 2027 Hospital Outpatient Prospective Payment System (OPPS) proposed rule, released July 7, 2026, CMS introduced a hopeful new idea: a payment category built specifically for this kind of technology, called Software as a Medical Service (SaMS), replacing the older, imprecise “Software as a Service.” The CY 2027 Physician Fee Schedule rule adopted the same definition days later, aligning both payment systems around one term for the first time.
Two concrete changes bring that idea to life:
- A new status indicator, “O1,” gives algorithm-driven services their own payment class for the first time. CMS proposes designating 36 existing codes accordingly.
- New Technology APC assignment gives these services a stable, separately payable rate of their own, rather than folding them into a broader clinical category where their value gets lost.
“As coding for SaMS continues to rapidly increase and technologies evolve, establishing a consistent and timely payment methodology for these services is a priority.” – CMS CY 2027 OPPS Proposed Rule (CMS-1850-P)
The Reasoning Matters as Much as the Mechanics
What makes this moment worth celebrating isn’t just the new codes — it’s the acknowledgment behind them. CMS said plainly that its outpatient payment logic, built for staff time and equipment, doesn’t fit software whose value comes from a trained model. It recognized that these technologies are arriving faster than its old categories could absorb, and that they deserved a framework of their own rather than an awkward fit into someone else’s.
For many medical device manufacturers, CMS is the benchmark for establishing payment. Its admission that its payment system is antiquated should be a welcoming signal for society: it offers hope for something better for everyone downstream of it, starting with patients.
What This Could Mean for Patients
Every one of those 83,392 claims in the JACR study represents a real person, at a real appointment, whose doctor believed an AI-powered tool was the right call. Multiply that across every specialty where this kind of technology is emerging, and the stakes come into focus: earlier detection, fewer invasive procedures, and care that reaches people sooner, wherever they happen to be seen.
Take heart failure. Rising pulmonary capillary wedge pressure typically precedes a hospitalization by days or weeks, but today, measuring it directly still means an invasive right heart catheterization in a procedural suite. Noninvasive technologies like Cardiosense’s CardioTag™ device exist precisely because that information shouldn’t require a procedure to obtain. Rather, it should be something a clinician can check easily and often, so that clinicians can serially track and manage their patients using the gold standard parameter in congestion assessment. A framework like SaMS is what turns technologies like this from a promising idea into something a patient’s own doctor can actually offer them.
That’s the real promise here: a future where the newest, smartest tools in medicine reach patients not years after they’re proven, but as soon as they’re ready. CMS took a real step toward that future this month, and it’s one worth celebrating.