Artificial intelligence is no longer a distant promise in American healthcare. Algorithms are already reading CT scans, chatbots are coaching patients through chronic disease, and wearable sensors are monitoring blood pressure between clinic visits. Yet for all this momentum, one fundamental question remains stubbornly unresolved: who pays for it?

In 2026, the Centers for Medicare and Medicaid Services (CMS) is taking its most ambitious steps yet toward answering that question; and the answers are illuminating both the promise and the peril of building a reimbursement infrastructure around clinical AI. Two developments in particular capture the tension: the launch of the ACCESS model, a sweeping experiment in outcomes-based payment for technology-enabled chronic care, and the establishment of a new Medicare payment code for AI-powered cardiovascular screening from routine CT scans. Together, they represent a paradigm shift in how the United States government thinks about paying for artificial intelligence in medicine.

"Participants will retain the right to determine if the structure of the model works for them and can choose to exit if it does not. Given the significant interest shown to date, we are confident in the design of the ACCESS Model." — Abe Sutton, Director, CMS Innovation Center

The ACCESS Model: Rewiring the Payment Logic for Digital Health

A New Experiment at Scale

In April 2026, CMS announced that more than 150 companies and providers had been provisionally approved to participate in the ACCESS model; the Advancing Chronic Care with Effective, Scalable Solutions initiative. The program, administered through the Center for Medicare and Medicaid Innovation (CMMI), was originally announced in late 2025 and is designed to pay participants set rates to manage chronic conditions including diabetes, hypertension, high cholesterol, musculoskeletal pain, anxiety, and depression. The first cohort launches in July 2026 and the model runs for ten years.

The list of accepted participants includes a who's who of the digital health world: mental health platforms Headspace, Jimini Health, Limbic, and Slingshot AI; cardiometabolic care companies Story Health, Cadence, and Verily (the Alphabet-backed life sciences firm); and wearable device makers Whoop and Withings. Weight management platform Noom is also among those accepted, signaling the breadth of digital health tools that may soon be operating within the Medicare system.

The volume of applications exceeded CMS officials' own expectations, and the agency extended the original April 1 deadline to allow more participants to apply. Notably, most applicants had not previously served Medicare patients — suggesting that ACCESS is opening a genuinely new market, not simply formalizing existing arrangements.

Outcome-Aligned Payments: The Core Innovation

The defining feature of ACCESS is its "outcome-aligned payment" structure. Rather than reimbursing for specific services or technology inputs — the logic that underpins the traditional fee-for-service system — ACCESS pays participants a bundled rate tied directly to measurable health outcomes. Payments range from a maximum of $180 to $420 per patient in the first year, depending on the condition treated, but participants earn the full amount only if at least half of their patients achieve defined clinical benchmarks. For hypertension, for example, patients are expected to lower systolic blood pressure below 130 mmHg, or reduce it by 15 mmHg.

This is a meaningful departure from the status quo. CMS has long struggled to reimburse digital health tools under the physician fee schedule, in part because its practice expense methodology was designed for physical medical equipment and clinical staff time — not software subscriptions, AI agents, or remote monitoring platforms. CMS acknowledged explicitly in its 2026 Physician Fee Schedule rulemaking that "innovative applications such as software algorithms and AI are not well accounted for" under existing payment frameworks, and solicited stakeholder input on how to incorporate SaaS and AI costs into future payment models.

The Human Clinician Question

ACCESS has no human clinician engagement requirement — a deliberate design choice that reflects where CMS thinks technology-enabled care is heading. Several participants are using this flexibility to rethink their operating models from the ground up.

Cadence, which currently helps hospitals manage patients with hypertension and other chronic conditions through remote patient monitoring codes that require monthly clinician contact, is undertaking what its CEO Chris Altchek describes as a wholesale redesign. The company is building out AI tools to handle tasks previously performed by human clinicians, with the goal of automating enough of the chronic care workflow that patients might see a clinician only once per year. As of early 2026, Altchek estimated Cadence had completed approximately 25% of the technology required to participate.

Story Health, which already relies heavily on automation to coordinate care between providers, nurses, coaches, and patients, is using ACCESS to accelerate its AI development roadmap. Chief Medical Officer Ashul Govil said the model has created "more of a reason to accelerate this effort" and advance AI features that were previously further out on the company's timeline.

"We don't really know what the upper limit of patient engagement is going to be. You can really only figure out what you have [AI] agents supporting these workflows in a way that it's very, very challenging to do with human clinicians." — Chris Altchek, CEO, Cadence

The Challenges: Rates, Rules, and Real Demand

Enthusiasm for the model is tempered by legitimate concerns. The payment rates announced in February 2026 disappointed many in the digital health community. The maximum $180 payment for behavioral health conditions — targeting anxiety and depression — was cited by Headspace CEO Tom Pickett as the reason his company had to abandon its original plan of partnering with a Medicare provider and instead enroll in Medicare directly. The rates, combined with a rule prohibiting ACCESS participants from billing Medicare for any other services delivered to their ACCESS patients, effectively rules out many integrated care providers.

Companies like PursueCare, a virtual substance use disorder provider with FDA-cleared software, chose not to participate at all. CEO Nicholas Mercadante stated that the billing exclusion was "problematic" and that "if there is little or no incentive — and possibly detriment — for a treater to participate, this won't be a very successful program."

Will Gordon, a former CMMI official now at Manatt, a legal and consulting firm, identified two fundamental questions that will determine the model's success over its ten-year life: whether Medicare beneficiaries can be engaged at scale through technology-enabled care, and whether there is real demand among Medicare's population for these kinds of care models. These are not rhetorical questions — they are genuine uncertainties that the experiment is designed to answer.

CMS plans to publish a list of approved companies along with their risk-adjusted clinical outcomes, creating a public accountability mechanism that will allow patients, providers, and policymakers to assess which technologies are actually delivering on their promises.

Opportunistic AI Screening: The Case of Coronary Calcium

A Window Into the Heart — and Into AI Reimbursement Policy

A different, and in some ways more targeted, window into the CMS reimbursement question opened on April 1, 2026, when Medicare established HCPCS code G0680 — a new payment pathway for AI-powered opportunistic coronary artery calcium (CAC) screening from routine chest CT scans. The payment rate is approximately $15 per scan, applying in the hospital outpatient setting.

The clinical logic is compelling. Coronary artery calcium — visible as bright, salt-like pixels in a CT image — is one of the most reliable predictors of future heart attack risk. Every year, roughly 19 million chest CT scans are performed in the United States, primarily to screen for lung cancer or evaluate persistent coughs. These scans routinely capture the heart, and an attentive radiologist can note any incidental calcium. But an estimated 20% to 40% of incidental calcium findings currently go unreported, leaving a large population with undetected cardiovascular risk.

AI algorithms from companies including Bunkerhill, Aidoc, Nanox, and HeartLung can automatically quantify coronary calcium and aortic valve calcification from existing chest CTs — without ordering additional scans, exposing patients to extra radiation, or requiring a radiologist to actively search for the findings. In March 2026, the American College of Cardiology and American Heart Association updated their guidelines to formally recommend the use of incidental calcium findings in guiding statin therapy, and for the first time explicitly invoked AI-based algorithms as a tool for automating that opportunistic screening.

Business Case vs. Evidence Base

For health systems, the appeal of AI-based opportunistic screening is partly clinical and partly financial. A Stanford trial published in 2023 found that patients whose CT scans flagged incidental CAC and who received a follow-up with their primary care physician were far more likely to start statin therapy than those who received usual care. But notified patients were also more likely to undergo follow-up cardiac testing — stress tests, coronary CT angiograms, and in some cases invasive coronary angiography. AI vendors estimate that for a large health system, the downstream care generated by a comprehensive opportunistic screening program could add tens of millions of dollars in new revenue.

This creates a structural tension that runs through much of clinical AI. Health informaticist Ken Mandl at Boston Children's Hospital has described the phenomenon as "biomarkup" — the tendency of health systems, consciously or not, to favor tests and technologies that lead to further testing, procedures, or prescribing, irrespective of whether those downstream interventions improve long-term outcomes. In the case of coronary calcium screening, the worry is that reimbursement, now established before robust long-term outcome data exists, could accelerate the deployment of algorithms whose ultimate benefit to patients remains unproven.

"We don't have the evidence that we should treat everything we find. If you don't have evidence that treating them can improve outcomes, you might be hurting people." — Morteza Naghavi, CEO, HeartLung AI

Adoption Barriers: The Reality Inside Health Systems

Despite the potential scale and the new reimbursement code, AI-based opportunistic CAC screening has been slow to take off. At Aidoc, one of the leading companies with an approved algorithm, just 20 of its 250 health system customers are currently using its CAC tool. The reasons illuminate the practical complexity of deploying AI in a real clinical environment.

Radiologists, who must typically double-check AI-generated calcium scores, are not separately compensated for that work — creating a classic misalignment between who bears the labor and who captures the revenue. Cardiology teams that are enthusiastic about the screening technology worry about being overwhelmed with referrals. At MedStar Georgetown University Hospital, cardiologists were initially hesitant to receive automated calcium scores for fear of flooding an already-backed-up practice with patients who had unexpected findings. In one health system using Bunkerhill's algorithm, 63,000 of 120,000 patients had moderate or severe calcium scores — but 85% of those were already being managed appropriately, meaning that identifying and routing the truly actionable 15% required additional clinical intelligence that the algorithm alone could not provide.

The $15 Medicare rate — lower than reimbursement for other cardiology algorithms — is designed partly as an experiment. Because G0680 is a temporary G code rather than a permanent CPT code, local Medicare Administrative Contractors (MACs) will make coverage determinations on a claim-by-claim basis, allowing CMS to accumulate real-world data before committing to a national coverage determination. As Jesse Ehrenfeld, Chief Medical Officer at Aidoc, noted: "That's what's great about a G code. They're temporary, they're flexible, they allow CMS to experiment with new models, because there are many unanswered questions."

The Broader Landscape: CMS and the AI Reimbursement Question

A Payment Infrastructure Built for the Past

The challenges illustrated by ACCESS and coronary calcium screening are not isolated — they reflect a systemic mismatch between the economics of AI and the architecture of American healthcare reimbursement. The Medicare Physician Fee Schedule, which governs how most outpatient services are paid, assigns relative value units (RVUs) to specific clinical procedures based largely on physician time, practice expenses, and malpractice costs. Software, AI inference, and data infrastructure do not map cleanly onto this framework.

CMS has acknowledged this limitation directly, noting in its 2026 PFS rulemaking that it "has been concerned about the rapidly changing nature of technology and the difficulty of obtaining verifiable and consistent costs from manufacturers." The agency is actively soliciting stakeholder input on how to incorporate SaaS and AI technology costs into evolving payment models — but the outcome of that solicitation has yet to translate into concrete reimbursement pathways at scale.

CMMI's Expanding AI Portfolio

ACCESS is not the only CMMI initiative with an AI dimension in 2026. The Wasteful and Inappropriate Service Reduction (WISeR) model, launched in January 2026 across six states, uses AI and machine learning — alongside human clinical review — to identify and reduce low-value services in Medicare. Technology companies participating in WISeR receive a percentage of expenditures associated with averted wasteful care, creating an incentive structure where AI is rewarded for reducing utilization rather than generating it. The contrast with the downstream revenue logic of opportunistic screening is instructive.

Meanwhile, the broader digital health ecosystem is evolving rapidly. CMS's Health Tech Ecosystem, launched in 2025 at the White House, aims to improve interoperability, expand patient-directed data mobility, and increase the availability of personalized digital tools for Medicare beneficiaries. More than 450 participants had joined the ecosystem by early 2026, preparing for initial platform releases.

State-Level Complexity: A Patchwork Emerging

While federal agencies move deliberately, states are legislating at a faster pace. By early 2026, 43 states had introduced over 240 AI-related bills — a volume nearly matching all of 2025. Common themes include transparency requirements when AI is used in clinical decisions, disclosure and opt-out mechanisms for patients, oversight requirements for AI use in prior authorization, and prohibitions on AI systems misrepresenting themselves as licensed clinicians. Colorado, Texas, Utah, and Illinois have enacted particularly detailed AI governance frameworks.

For health AI companies operating nationally, this creates a compliance environment of considerable complexity. A tool approved by the FDA, reimbursed by CMS, and deployed in a hospital in Texas must simultaneously comply with Texas's AI disclosure requirements (effective January 2026), California's chatbot guardrails (effective January 2026), and Colorado's AI impact assessment mandates (enforcement beginning June 2026). The federal reimbursement question is only one layer of a multi-jurisdictional regulatory stack.

Implications for Health AI Companies

Design for Reimbursement from Day One

The ACCESS model and the coronary calcium G code both demonstrate that CMS is increasingly willing to create novel payment pathways for AI — but that those pathways are structured around specific conditions, defined outcomes, and verifiable data. Companies building clinical AI tools should design their evidence and data collection strategies with reimbursement requirements in mind from the earliest stages of development, rather than treating payment as a downstream commercial consideration.

Outcome Evidence Is Non-Negotiable

The clearest lesson from the coronary calcium debate is that surrogate endpoints — getting more patients on statins, improving cholesterol scores — are insufficient to establish the long-term value of AI interventions. For reimbursement to be sustainable, companies need randomized or large-scale observational evidence demonstrating that AI-detected findings translate into fewer heart attacks, strokes, hospitalizations, and deaths. Until that evidence exists, regulators, payers, and clinicians will remain appropriately cautious about broad deployment — as they should be.

Think Beyond Fee-for-Service

ACCESS signals that the most viable near-term reimbursement pathways for digital health AI may lie outside traditional fee-for-service models entirely. Outcome-aligned, value-based, and risk-sharing payment arrangements — where the technology company or provider bears some of the financial risk associated with clinical performance — may better reflect the economics of scalable AI than attempting to fit algorithms into legacy billing codes. Companies that can demonstrate consistent, measurable health improvements across a defined patient population will be best positioned to thrive in this environment.

Address Workflow Integration Honestly

The slow uptake of coronary calcium AI at major health systems is a reminder that even technically superior tools fail if they create uncompensated work, disrupt clinical workflows, or overwhelm care teams. The radiologist who must double-check an AI score without additional pay; the cardiologist facing a flood of unexpected referrals; the primary care physician receiving an automated alert without context or support — these are not edge cases, they are predictable consequences of deploying population-scale AI into a system built for individual patient encounters. Successful commercialization requires genuine workflow integration, not just technical performance.

Conclusion: Paying for Progress

The question of CMS reimbursement for AI is, at its core, a question about what kind of healthcare system the United States wants to build. The ACCESS model represents a bet that technology — and specifically AI — can manage chronic disease at scale, at lower cost, and with measurable health benefits for tens of millions of Medicare beneficiaries. The coronary calcium screening pathway represents a bet that the same logic extends to preventive screening: that AI can find the patients most at risk before they show up in emergency departments, and that catching them earlier will ultimately cost less and save more lives.

Both bets may prove correct. The evidence base is growing, and the institutional commitment from CMS is real. But the history of healthcare technology adoption counsels humility. Reimbursement is necessary but not sufficient. Genuine clinical benefit, equitable access, careful surveillance of unintended consequences, and honest accounting for the full costs — financial, clinical, and human — of deploying AI at scale are equally essential.

At gesund.ai, we believe that navigating this landscape successfully requires a clear-eyed understanding of both the opportunities and the obligations that come with building AI for healthcare. The payment infrastructure is evolving rapidly. The science must keep pace.

For inquires, please contact us.

Bibliography

1. Aguilar, M. (2026, April 13). Access granted: CMS greenlights more than 150 participants for chronic care experiment. STAT News.

2. Palmer, K. (2026, April 15). AI could check millions of CT scans for heart risk. Who will pay for it? STAT News.

3. CMS Innovation Center. (2025, December). Improving ACCESS to technology-supported care with outcome-aligned payments. Centers for Medicare and Medicaid Services. cms.gov.

4. HeartLung AI. (2026, April 7). HeartLung AI welcomes CMS coverage for AI-enabled opportunistic coronary artery calcium (CAC) detection on chest CT. Press release. heartlung.ai.

5. Healthcare Dive. (2026, April 13). CMS accepts more than 150 providers, digital health firms for ACCESS model. healthcaredive.com.

6. Nixon Peabody. (2025, December 3). CMS announces new value-based payment model for technology-enabled care. nixonpeabody.com.

7. Sidley Austin LLP. (2025, December 10). CMS finalizes key Medicare reimbursement policies for telehealth, chronic disease, and digital health. sidley.com.

8. McDermott+. (2025, July 31). Back for seconds: digital health policies in the CY 2026 PFS proposed rule. mcdermottplus.com.

9. Nixon Law Group. (2025, September 12). Digital health stakeholders: weigh in on CMS 2026 chronic disease and virtual care RFIs. nixonlawgroup.com.

10. Centers for Medicare and Medicaid Services. (2026). Wasteful and Inappropriate Service Reduction (WISeR) Model. cms.gov.

11. Manatt Health. (2026, April). Health AI policy tracker. manatt.com.

12. blueBriX. (2026, January 29). The 2026 AI reset: a new era for healthcare policy. bluebrix.health.

13. American College of Cardiology / American Heart Association. (2026, March). Updated guideline on cholesterol management and AI-based coronary calcium screening.

14. Sandhu, A. et al. (2023). AI-based incidental coronary artery calcium detection and statin initiation: A randomized trial. Stanford University / Bunkerhill Health.