CMS's ACCESS model launches in July. It is the first time Medicare has tried to pay for AI-managed chronic care at scale.

On 5 July 2026, the CMS Innovation Center will start the first performance period of the Advancing Chronic Care with Effective, Scalable Solutions (ACCESS) model — a ten-year voluntary Medicare payment model designed around what CMS is calling “Outcome-Aligned Payments.”

The acronyms and the duration tell you something. ACCESS is the longest-running CMMI model launched in recent memory, and the Outcome-Aligned Payment mechanism is the first time CMS has explicitly built a payment structure around the assumption that technology — including AI diagnostics — will be doing some of the clinical work that previously required a billed clinical encounter.

The model deserves more attention than it has received, and not for the usual reason payment-reform stories matter. ACCESS is the closest thing yet to a real federal price for the value AI is supposed to deliver in chronic care.

What ACCESS actually pays for

The model targets four conditions that, together, affect more than two-thirds of Medicare beneficiaries: hypertension, type 2 diabetes, chronic musculoskeletal pain, and depression.

Participating organizations — health systems, ACOs, primary-care groups, and chronic-care management vendors — receive predictable per-beneficiary monthly payments to manage one or more of these conditions across a panel of Original Medicare patients. The structure is roughly familiar from earlier CMMI primary-care models. What is new is the back-end.

The full payment is earned only when patients meet measurable health goals — for hypertension, that means documented blood-pressure control; for diabetes, glycemic control above a defined threshold; for the other two conditions, similar outcome bundles. Participants who enroll patients but do not move the outcomes do not earn the full per-member payment. CMS retains the difference.

This is not unique as a value-based-care design idea. What is unusual is that CMS has, in the model’s own announcement, explicitly named the technology layer it expects participants to build: “artificial intelligence diagnostics that identify people with conditions that might benefit from ACCESS services, devices that monitor biomarkers, and software that streamlines key workflows.”

CMS does not write that kind of sentence in a model announcement by accident. It is signalling, to vendors and to investors, what the model expects.

Why this is a different kind of model

There are two technical reasons ACCESS is more interesting than the usual chronic-care payment reform.

First, AI diagnostics get paid for finding the patient, not just managing them. Most prior CMS efforts to support AI in care delivery have focused on the diagnostic encounter — the AI-augmented scan, the algorithm-flagged ED case. ACCESS is the first model that effectively pays for cohort identification at scale: AI tools that surface previously uncontrolled hypertensives in a primary-care panel, that risk-stratify diabetics for outreach, that find depression in undertreated populations.

This matters because identification has historically been the unfunded step. A primary-care practice can run an algorithm against its panel and surface 200 patients with uncontrolled hypertension — and then, under fee-for-service, has no reimbursement pathway for the outreach and management work that produces an actual outcome on those patients. ACCESS is designed to close that gap.

Second, biomarker monitoring and software become reimbursable substrate, not billable encounters. Under the existing Chronic Care Management and Remote Patient Monitoring codes, the reimbursable unit is clinician time tied to a specific patient. Under ACCESS, the unit shifts to outcome achieved per panel member. That decouples the value of the software from the clinician minute, which is the right direction if you believe — as the model implicitly does — that the next decade of chronic-care productivity gains will come from algorithm-driven population management rather than from more billed minutes of clinician attention.

What this does to the vendor market

The implications for the technology vendor market are sharper than the press release suggests.

Vendors selling RPM devices, chronic-care apps, and AI population-health tools have spent the last five years pitching to provider organizations on the basis that their tools generate billable activity. ACCESS shifts the value proposition. The question a participating provider organization will now ask its vendors is: do you measurably move the outcome that we get paid for?

The vendors who answer that with peer-reviewed data, large real-world studies, or transparent risk-adjusted outcomes will be the ones that win ACCESS contracts. The vendors whose evidence base is a slide deck and a 30-patient pilot will not.

This is the kind of pressure the AI-in-healthcare market has badly needed. The current crop of healthcare LLM and population-health vendors is over-supplied with marketing claims and under-supplied with evidence. ACCESS is, structurally, an evidence-forcing function.

It is also a winnowing function. Vendors that do not have plausible outcomes data by the end of the first performance period will struggle to renew contracts into the second.

Where it is likely to disappoint

A few honest caveats.

CMMI models have a long track record of producing modest savings and uneven adoption. ACCESS is voluntary, which means it will attract organizations that already think they can hit the outcomes — selection bias that limits how much it tells us about the underlying technology’s effectiveness in average settings.

The condition list is narrow enough that the model will probably not be the answer for the more complex chronic-care problems where AI’s clinical case is strongest: multimorbidity, frailty, transitions of care. Those remain in policy limbo for now.

And ACCESS’s ten-year horizon, while genuinely useful for vendor investment, also exposes the model to political risk — administration changes, CMMI restructuring, and broader Medicare reimbursement debates over the same decade.

The signal in the noise

For all those caveats, ACCESS is the most concrete sign yet that CMS has begun to design payment infrastructure around the assumption that AI will be doing real chronic-care work. The model is small relative to total Medicare spend. It is large relative to every prior CMS attempt to price clinical AI directly.

The 5 July 2026 launch date matters less than the second-period applications window, which determines whether the first cohort’s results pull additional participants in or scare them off. The first reasonable read on whether ACCESS works will arrive in late 2027.

Until then, it is the right place for any AI-in-chronic-care vendor or provider organization to be reading their incentives from.