Medicare Will Pay AI Agents for Chronic Care Starting July 2026 — And Most of Tech Hasn't Noticed

While Washington debates sweeping AI legislation and Silicon Valley debates who has the fastest model, a quieter and potentially far more impactful policy change is about to take effect: starting July 5, 2026, Medicare will pay for AI.

Not for the privilege of having an AI assistant in a doctor’s office. Not for generating clinical notes faster. For the first time in the program’s 60-year history, the Centers for Medicare & Medicaid Services will reimburse health technology organizations based on whether their AI-powered tools actually improve patient health outcomes — opening a reimbursement pathway that has never existed for the class of digital chronic-care management systems that health startups have been building for years without a viable payment model.

The program is called ACCESS: Advancing Chronic Care with Effective, Scalable Solutions. It may be the most consequential federal AI policy development of 2026 that nobody in the tech industry is talking about.

The Problem Access Is Trying to Solve

Medicare’s historical payment structure rewards volume — office visits, procedures, lab tests — not outcomes. A diabetic patient who misses three medication refills, skips a follow-up appointment, and ends up in the emergency room generates more Medicare revenue than a patient who, aided by daily check-in calls from a care agent, maintains their blood sugar within target range and stays out of the hospital.

Digital health companies have spent a decade building tools that work like the second scenario: apps that monitor patients between visits, algorithms that flag deteriorating trends before they become crises, voice agents that call patients to confirm they filled a prescription or ask about a change in their energy levels. The tools work. The problem has been that there was no mechanism to pay for them within Medicare.

That mechanism now exists.

How ACCESS Works

The ACCESS model is a voluntary 10-year pilot administered by the CMS Innovation Center. Participating organizations — digital health companies, provider groups, and integrated care systems — apply to manage specific Medicare populations with qualifying chronic conditions. They receive predictable per-patient monthly payments for enrolling beneficiaries and managing their care through technology-supported protocols.

The twist is that the full payment is contingent on outcomes. For the 2026-2027 performance period, at least 50% of a participant’s enrolled beneficiaries must meet all required outcome targets for the organization to receive complete reimbursement. The targets are condition-specific: a hypertensive patient who lowers their blood pressure by 10 mmHg, or a diabetic patient who reduces their HbA1c by a clinically meaningful margin, counts toward the threshold. Enrolling a patient and doing nothing with them does not.

The model covers four clinical tracks that together affect approximately two-thirds of all Medicare beneficiaries. The first two tracks — early cardio-kidney-metabolic (covering hypertension, dyslipidemia, obesity, and prediabetes) and full cardio-kidney-metabolic (covering type 2 diabetes, chronic kidney disease, and established cardiovascular disease) — target the most expensive cluster of conditions in the program. The remaining tracks address musculoskeletal pain and behavioral health.

What the AI Actually Does

The care delivery model that ACCESS was designed to accommodate is what the healthcare industry calls “technology-supported chronic care” — services delivered largely or entirely outside the clinic walls, between scheduled appointments, by AI agents and care coordinators rather than physicians.

In practice, this means: AI agents that conduct regular wellness check-ins by phone or text, flagging patients who report new symptoms or missed medications for human follow-up. Algorithms that parse biometric data from wearables and connected devices to identify patients trending toward a crisis before the patient notices anything wrong. Care navigation tools that automatically schedule follow-up appointments, identify social determinants of health — like whether a patient has reliable transportation to a pharmacy — and connect patients with community resources.

None of these services could previously be billed to Medicare in a meaningful way. A physician’s office visit could be reimbursed; a voice agent calling a patient on a Tuesday afternoon could not, even if that call demonstrably prevented a hospitalization that would have cost Medicare tens of thousands of dollars.

ACCESS breaks that logic by paying for the outcome rather than the individual service. The method by which a participating organization achieves the outcome — whether through frequent AI-driven check-ins, intensive human case management, or a combination — is left to the organization.

150 Companies, 350 Applicants

More than 350 technology-enabled care organizations submitted letters of intent to apply for the program. CMS selected 150 for the first cohort, which begins operations July 5, 2026.

The participating companies span a range of approaches. Some are large integrated health systems that have built internal AI platforms. Others are pure-play digital health startups that contract with provider groups to manage specific patient populations. The cohort includes companies using large language models for patient communication, computer vision for remote diagnostic monitoring, and specialized narrow AI for condition-specific risk stratification.

The application deadline for the first performance period passed on May 15. Organizations that missed the cutoff can apply for a January 1, 2027 start date.

Why This Matters Beyond Healthcare

The ACCESS model represents something the AI industry has struggled to achieve in other sectors: a federal commitment to paying for AI based on verified outcomes, at scale, in a domain that is both economically enormous and societally critical.

Medicare spends approximately $1 trillion per year. Chronic conditions — the ones ACCESS targets — account for roughly 80% of that spending. If outcome-based payments for AI chronic care management succeed at even modest scale, they will generate more real-world evidence about AI efficacy in high-stakes decision environments than any benchmark or academic study.

The model also creates competitive pressure that could reshape the digital health industry. Companies that can demonstrate measurable outcomes will be paid. Companies that cannot will not. This is a more rigorous test of AI product quality than anything the private market has imposed — and the consequences of failing are measured in patient health, not quarterly earnings.

For technology investors and builders, the practical implications are significant. A viable Medicare payment pathway transforms the addressable market for AI chronic care tools from a small slice of employer-sponsored health plans and direct-to-consumer subscriptions into the largest single payer in the American healthcare system. The revenue model for AI-assisted chronic care has fundamentally changed.

The Regulatory Question

CMS structured ACCESS as a pilot under the CMS Innovation Center’s existing statutory authority, which means it does not require new legislation. The Innovation Center has broad latitude to test payment and delivery models, and if ACCESS demonstrates savings and improved outcomes, the agency can move to expand it to more beneficiaries or make it a permanent part of Medicare payment policy — again, without an act of Congress.

That pathway makes ACCESS unusual among federal AI policy initiatives. While the proposed America AI Act remains contested and the EU AI Act’s high-risk provisions are still phasing in, ACCESS is already built, funded, and set to launch in five weeks with a defined set of participants, a clear payment structure, and measurable success criteria.

Whether it works is the open question. Outcome-based payment models have a checkered history in Medicare — some have generated savings, others have been discontinued after inconclusive results. The specific challenge for AI-powered chronic care is that demonstrating causality is hard: if a patient improves, was it because of the AI check-in program, the new medication their doctor prescribed last November, or a behavioral change they made independently?

ACCESS sidesteps this problem by measuring outcomes at the population level rather than the individual case level. If a participating organization’s cohort of hypertensive patients, in aggregate, achieves better blood pressure control than a matched Medicare population, the organization gets paid. Whether any individual improvement was AI-caused is not the question being asked.

That is either a pragmatic design choice or a significant methodological limitation, depending on who you ask. What it is not, for the first time, is theoretical.

The AI is in the building. Medicare is going to pay for it. On July 5, we find out what it can actually do.