Medicare Just Created a Payment Pathway for AI Agents — and Silicon Valley Barely Noticed
On December 1, 2025, the Centers for Medicare & Medicaid Services quietly announced a 10-year pilot called the ACCESS model — Advancing Chronic Care with Effective, Scalable Solutions — and with it did something Medicare has never done before: created a formal payment mechanism for AI agents operating between clinical visits. No doctor required for the AI check-in. No face-to-face time billed. Just outcomes.
The model goes live July 5, 2026. More than 150 organizations — digital health startups, AI doctor companies, wearable makers, and a Google-backed life sciences firm — have already been accepted into the first cohort. They will receive recurring monthly payments for managing chronic conditions in traditional Medicare patients, with full payment contingent on measurable health improvements. The AI that calls a diabetic patient to confirm she took her metformin, or routes a hypertensive patient to a housing-stability program, can now be the billable service.
Medicare is a $900 billion program. When it decides something is reimbursable, private insurers typically follow within 18 to 24 months. The AI industry spent the past two years debating which chatbot benchmark matters. CMS spent that time building the reimbursement rails that will determine which AI companies get paid at federal scale — and most of Silicon Valley has no idea the tracks were laid.
- ~44Mpotential Medicare FFS beneficiariesACCESS targets Original Medicare (fee-for-service) patients with at least one of four covered chronic condition clusters — roughly two-thirds of all Medicare enrollees. Source: CMS.gov ACCESS model documentation.
- $900Bannual Medicare program spendMedicare sets the reimbursement standard the entire U.S. healthcare system tracks. Private insurers follow Medicare payment decisions within 18–24 months. Source: TechCrunch, May 2026.
- 10 yearsmodel performance periodJuly 5, 2026 through June 30, 2036. Rolling applications open through April 1, 2033. Source: CMS CMMI ACCESS model page.
- 150+accepted first-cohort organizationsAccepted as of April 2026, exceeding CMS expectations. Most had never previously served Medicare beneficiaries. Source: STAT News, April 13, 2026; CMS Accepted Applicants list.
- $7.50–$35per-beneficiary per-month payment rangeRates contingent on clinical outcome achievement. Co-management with primary care physicians: ~$30 per review, roughly $100 annually per patient. Source: Healthcare Dive, April 2026; Foley Hoag analysis.
ACCESS is not an insurance product. It is not a Medicare Advantage plan. It is a payment model test run by the CMS Innovation Center (CMMI) under its authority to test alternative payment approaches in traditional Medicare. Participating organizations — which can be clinics, health tech companies contracting with enrolled Medicare providers, or hybrid entities — receive monthly recurring payments for managing patients’ chronic conditions across four clinical tracks.
Track 1 — Early Cardiometabolic & Kidney (eCKM): Hypertension, dyslipidemia (high cholesterol), overweight/obesity, prediabetes. Outcome measures: blood pressure control, lipid panels, HbA1c, weight. Target population: the largest single chronic-disease cluster in Medicare.
Track 2 — Late Cardiometabolic & Kidney (CKM): Diabetes, chronic kidney disease, established cardiovascular disease. Measures: BP, lipid control, HbA1c, eGFR/UACR reporting. The highest-cost cohort in Medicare — diabetes and CKD alone drive disproportionate per-capita spending.
Track 3 — Musculoskeletal (MSK): Chronic musculoskeletal pain. Measures: patient-reported pain intensity, physical function (PROMs). No prior-authorization required for MSK digital tools under ACCESS rules.
Track 4 — Behavioral Health (BH): Depression, anxiety. Measures: PHQ-9 and GAD-7 score improvement, WHO Disability Assessment 2.0 (WHODAS 2.0) functional assessments. The first time behavioral health apps have had a clear Medicare payment track for outcome-based reimbursement.
The defining innovation is the payment structure: Outcome-Aligned Payments (OAPs). Traditional Medicare pays for clinician time, procedures, and face-to-face encounters. ACCESS pays for outcomes — and critically, does not specify how you achieve them. A software company, an AI voice agent, a remote monitoring platform, or a human care coordinator all qualify, as long as the patient’s blood pressure drops.
The phrase “built for AI” requires unpacking, because ACCESS does not mention artificial intelligence by name in its clinical requirements. What it does instead is more powerful: it removes all the friction points that have kept AI out of Medicare reimbursement for a decade.
1. Outcome-not-activity payment: Traditional Medicare requires billing codes tied to specific clinician actions — a CPT code for a 15-minute office visit, a G-code for a remote monitoring session. ACCESS pays for outcomes and does not mandate the activity that produces them. An AI agent that autonomously monitors a patient’s blood pressure via a connected cuff, alerts a care team to a spike, and coordinates a medication adjustment is the billable service — even though it never involved a scheduled visit.
2. Technology-agnostic care delivery: CMS explicitly permits care delivered through “in-person visits, virtual consultations, asynchronous communication, or technology-enabled methods,” as well as “use or monitoring of FDA-authorized devices or software.” Asynchronous AI-to-patient communication — the model that operates at lowest cost — is explicitly on the approved list.
3. Non-clinician organization eligibility: For the first time, health tech companies that are not themselves Medicare providers can participate by contracting with enrolled clinicians to serve as Clinical Directors. A startup with zero Medicare billing history can receive ACCESS payments if a licensed physician supervises the clinical arm. This is the carve-in the AI industry needed.
4. Cost-sharing waiver authority: ACCESS participants may completely waive patient cost-sharing for enrolled beneficiaries — an inducement that traditional Medicare prohibits. This lets AI-first companies acquire patients without the premium/copay friction that has stalled digital health adoption in Medicare Advantage, and without the competitive disadvantage of conventional billing.
“If you want to build a model that truly incentivizes the use of AI, the reimbursement system has to stop asking what you did and start asking what changed.”
Neil Batlivala, CEO — Pair Team · TechCrunch, May 12, 2026
Pair Team — one of the 150 accepted participants — illustrates how this plays out in practice. The company manages roughly 500,000 patients facing chronic conditions compounded by social barriers: housing instability, food insecurity, lack of transportation. It uses a voice AI agent named Flora for intake, referral coordination, and patient engagement — tasks that in the traditional fee-for-service model could only be billed if a clinician conducted them personally. Under ACCESS, Flora’s work is the service. The clinical outcome is the bill.
ACCESS is the product of a specific bet made by two former technology-sector officials who now run CMMI. Understanding who they are explains why a Medicare payment model ended up structured this way.
Abe Sutton — CMMI Director: Sutton came to CMS from Rubicon Founders, a healthcare venture firm. As CMMI’s director he has explicitly pushed for more mandatory payment models and told reporters the innovation center must “get more healthcare providers, especially poor-performing ones, to participate in value-based care.” On data sharing, he told the AMA directly: “we’re building in requirements for the organizations participating in this model, that they share back that data.”
Jacob Shiff — Chief AI & Technology Officer, CMMI: Former healthcare founder who joined the Innovation Center to embed AI-readiness into CMMI’s model design. ACCESS is the first model with a formal CMMI Chief AI Officer fingerprint on its payment structure.
Dr. Mehmet Oz — CMS Administrator: Oz publicly endorsed the model on December 1, 2025, appeared in a two-minute YouTube explainer alongside Sutton, and has framed digital health and AI as core to his “Make America Healthy Again” agenda at CMS. He told the AMA House of Delegates: “Physicians have lost their mojo” — and positioned AI tools, including the WISeR prior authorization model, as a way to get it back.
RFK Jr. — HHS Secretary (oversight): The ACCESS model sits under HHS Secretary Robert F. Kennedy Jr.’s MAHA health agenda, which emphasizes chronic disease prevention and technology-enabled care as alternatives to pharmaceutical-first treatment pathways.
The December 2025 launch came with a parallel track of infrastructure: CMS also published the TEMPO (Technology-Enabled Meaningful Patient Outcomes) pilot in the Federal Register the same week, establishing FDA-device integration standards that complement ACCESS. On July 30, 2025, President Trump, Kennedy, and Oz held a White House event to launch the CMS Interoperability Framework, gathering voluntary commitments from EHR vendors, health information networks, and technology platforms to implement standardized data exchange — the data plumbing that makes outcome-tracking at ACCESS scale feasible.
When CMS opened applications in January 2026, the response exceeded expectations. By April, 150 organizations had been provisionally accepted. Most had never billed Medicare. That is a significant fact: the model is drawing in an entirely new class of healthcare provider, and the incumbent fee-for-service ecosystem is watching from the outside.
Noom(weight management / behavioral change) — Track 1/2 (eCKM/CKM). Enrolls patients with obesity and metabolic conditions using psychology-based digital coaching. First time Noom has a Medicare payment pathway.
Verily(Alphabet-backed life sciences) — Metabolic tracks. Brings Google-scale data infrastructure and remote monitoring hardware to Medicare for the first time under this structure.
Whoop(wearable) — Connected biometrics. Heart rate, sleep, recovery data fed into clinical dashboards for outcome tracking.
Doctronic(AI doctor startup) — LLM-based clinical workflows. One of the most direct bets that ACCESS will pay for AI-native care delivery.
Berry Street(virtual nutrition therapy) — Track 1/2. Registered dietitians plus digital coaching; metabolic outcomes reimbursed under ACCESS rather than requiring individual credentialed-visit billing.
Devoted Medical(clinical arm of Devoted Health) — Metabolic tracks. Notable: Devoted is a Medicare Advantage plan, but ACCESS is an Original Medicare program — Devoted Medical is participating separately, as a clinical organization, in fee-for-service.
Aledade & Guidehealth(ACO enablement) — Accountable Care Organization support organizations bringing AI-assisted risk stratification to primary care networks enrolled in ACCESS.
Pair Team(complex chronic + social determinants) — Flora AI voice agent. 500,000 current patients; targeting 1 million within three years under ACCESS reimbursement. Funded by Kleiner Perkins.
Withings(connected devices) — Blood pressure cuffs, scales, sleep trackers feeding FHIR-compliant data streams to participating care organizations.
The participant list reveals a strategic split. On one side: AI-native startups (Doctronic, Pair Team) betting the payment model will validate AI-first care delivery at federal scale. On the other: established health systems and ACO enablers (Aledade) using ACCESS to layer AI tools onto existing Medicare networks. And in between: consumer-facing tech companies (Noom, Whoop, Withings) converting their retail hardware or app subscriptions into Medicare-billable clinical services for the first time.
The model’s equity provisions are written into the outcome measures themselves — ACCESS pays on directional improvement from each patient’s individual baseline, rather than on reaching a universal target. A patient whose blood pressure starts at 180/110 and drops to 155/95 is credited with the same category of improvement as a patient who goes from 145/90 to 130/80. In theory, this should prevent the model from rewarding organizations that cherry-pick already-healthy or already-compliant patients.
CMMI’s own prior model reviews have found implicit bias baked into risk-adjustment and screening tools. An analysis of three previous payment models — Kidney Care Choices, Comprehensive Care for Joint Replacement, and Million Hearts Cardiovascular Disease Risk Reduction — found that risk-scoring algorithms embedded in those models systematically underestimated clinical need for Black and Hispanic patients.
ACCESS’s AI tools, most of which will be built and owned by participating vendors rather than CMS, will inherit whatever training-data biases exist in the underlying models. A voice agent trained primarily on English-language, high-health-literacy patient interactions will perform differently on Spanish-speaking elderly patients with limited education. The outcome measures will capture the gap — but only after the fact.
HHS Section 1557 nondiscrimination rules (updated May 2024) now explicitly ban discrimination by AI-based clinical decision tools, with compliance required by May 2025. ACCESS participants are subject to those rules. Whether CMS will audit AI vendor bias proactively — or only after outcomes data shows disparities — is unspecified in the model design.
A 2025 peer-reviewed scoping review in Frontiers in Public Health found that AI models trained on historically under-resourced patient populations carry documented systematic bias, and that in low-resource settings the harm from biased AI tools can exceed the benefit from any efficiency gains.
“Outcome-based payment sounds like the solution to cherry-picking. But if the AI tool performs worse on your population, you miss the outcome, lose the payment, and exit the market — which is the same result as cherry-picking, just slower.”
Analysis — Frontiers in Public Health · Algorithmic Bias in Public Health AI, 2025
The AI industry’s fixation on benchmark leaderboards and GPU scarcity has left it largely unaware of the slowest-moving but most durable transformation in AI deployment: what the federal government agrees to pay for. Medicare is a $900 billion annual program. Its coverage decisions are the single most powerful signal in U.S. healthcare economics. When Medicare decides AI agents can bill for patient monitoring, medication adherence, and housing-referral coordination, every private insurer in the country receives a 12-to-24-month countdown clock to match that coverage or lose employer contracts.
The prior authorization channel: In parallel with ACCESS, CMS is implementing the Interoperability and Prior Authorization Final Rule (CMS-0057-F), which mandates FHIR-based Prior Authorization APIs from all major payers by January 1, 2027. CMS projects $15 billion in 10-year savings from this automation. Every AI company building prior-authorization workflow tools — Suki, Abridge, Notable Health, Nabla, and dozens of others not yet in ACCESS — is building for a market that CMS is mandating into existence.
The EHR vendor pressure: Epic and Oracle Health (Cerner) have already integrated ambient AI documentation tools (Nuance DAX, Abridge). ACCESS adds a new pressure: organizations that want to participate must demonstrate data sharing with primary care physicians — which means their AI tools must write back to the EHR in FHIR-compliant formats. The EHR vendors who gatekeep those data flows become chokepoints, or partners, in every ACCESS deal.
The CBO headwind: Between 2011 and 2024, CMMI scaled only four of the 70 models it tested. A 2023 CBO assessment found CMMI increased federal spending by $5.4 billion over its first decade rather than achieving net savings. Congress is watching. ACCESS has a data-publication requirement — annual outcome reports — and CMS has committed to recalibration if results underperform. Companies that build their Medicare revenue model around ACCESS payments are exposed to model termination risk.
None of this shows up on a SWE-bench leaderboard. It does not generate a product announcement or a demo reel. But the companies inside those 150 approved slots — and the EHR vendors, AI scribes, connected-device makers, and prior-authorization platforms that plug into them — are operating in the first federally funded AI deployment infrastructure the U.S. health system has ever created. The benchmark that matters here is not accuracy on a coding task. It is whether a hypertensive 72-year-old in rural Georgia gets her blood pressure under 130 by January.
The ACCESS model — Advancing Chronic Care with Effective, Scalable Solutions — is the first federal payment mechanism designed to reimburse AI-driven chronic care at Medicare scale. It launches July 5, 2026, with 150+ participants and a 10-year runway. CMS is a $900 billion price-setter. When it pays for AI agents, the private insurance market follows. Silicon Valley is still benchmarking chatbots. The healthcare AI land grab is happening inside Medicare, right now.
STARTING NOW — Join CMS virtually for a special event exploring the new ACCESS Model and how it will help people access tech to take charge of their health. Join now.
ViVE 2026: CMMI’s Abe Sutton shares insight into ACCESS Model.
Dr. Mehmet Oz, Administrator of the Centers for Medicare & Medicaid Services (CMS): “You own your medical records. You paid for them one way or another, and we’re making sure you can give those medical records to whoever you want. For the first time, we can make it easier for AI…”
Today, we’re announcing the members of our new Federal Healthcare Advisory Committee, which will develop recommendations for how @SecKennedy and I can improve Medicare, Medicaid, the Children’s Health Insurance Program, and the Health Insurance Marketplace.
Today, @SecKennedy and @DrOzCMS met with industry leaders to discuss their pledge to reform and improve the prior authorization processes for Medicare Advantage, Medicaid Managed Care, Health Insurance Marketplace and commercial plans covering nearly eight out of 10 Americans.
HHS Secretary Robert F. Kennedy Jr. — Truth Social — on the Make America Healthy Again agenda that frames CMS’s December 2025 launch of the ACCESS chronic-care model.