Thursday, June 25, 2026

The June 25 HHS Update on AI in Clinical Care: Meeting Summary

In December 2025, HHS announced an RFI on how it could improve access to AI for clinical care.  See webpage here, Fed Reg here.  (90 FR 60108, 12/23/2025).

On June 25, HHS held an hour-long webinar to update the public on progress.  While the three arms of the program are regulation, research and development, and reimbursement, they opened by announcing this webinar would include regulation and R&D (not reimbursement).

Claude Opus summarizes the transcript.

###

HHS Hosts a "One HHS" Clinical-AI Webinar — A Landscape Tour, Not a Comment Readout

A working record of the June 2026 HHS webinar tied to the December 23, 2025 Request for Information, "Accelerating the Adoption and Use of Artificial Intelligence as Part of Clinical Care" (RIN 0955-AA13).

Summary  

In June 2026, HHS held a public webinar tied to its December 23, 2025 Request for Information on accelerating AI in clinical care (RIN 0955-AA13). Rather than reporting comment feedback, seven officials sketched a landscape of departmental AI activity under Secretary Kennedy's "One HHS" banner. The RFI names three levers—regulation, reimbursement, and research & development—but the webinar addressed only regulation and R&D; reimbursement was noted to be out-of-scope, for today.

Speakers positioned AI as a "third lever" alongside lifestyle and care infrastructure within the Make America Healthy Again agenda, emphasizing cost growth, chronic-disease management, agentic clinical AI (ARPA-H's ADVOCATE), caregiving and aging (ACL), and risk-proportionate FDA oversight.

Orientation: what this webinar was, and wasn't

The December 2025 RFI (published at 90 FR 60108, Dec. 23, 2025; comments closed February 23, 2026; signed by Secretary Robert F. Kennedy, Jr.) sits downstream of the HHS AI Strategy issued December 4, 2025, OMB Memorandum M-25-21, and the July 2025 America's AI Action Plan. It solicits input across three approaches—regulation, reimbursement, and research & development—and poses ten specific questions ranging from barriers to private-sector innovation to evaluation methods and interoperability.

Two things are worth flagging up front for readers who came expecting substance on the RFI itself:

  1. This was not a comment readout. No analysis of submitted feedback was presented. The closest the webinar came was the Deputy Chief AI Officer's high-level recitation of "three needs we heard" (coordination, adoption support, evaluation/benchmarking). The body of the event was a sequence of brief division-by-division spotlights.

  2. Reimbursement—one of the three named levers—was effectively skipped. It was acknowledged in passing as something HHS "does," and cost growth framed much of the discussion, but no payment-policy direction was offered. For an audience that lives in coverage and payment, the absence of the lever most proximate to diagnostics and laboratory reimbursement is the most conspicuous feature of the agenda.

What follows is a speaker-by-speaker record.

1. AI as MAHA's Third Lever — Dr. Thomas Keane

Assistant Secretary for Technology Policy / National Coordinator (ASTP/ONC)

Keane, an interventional radiologist, opened with a clinical anecdote: activating a stroke network so that catheter-based clot retrieval (mechanical thrombectomy) could restore a patient who, in an earlier era, would not have been treatable. His point was that AI now sits inside care that did not exist when he trained—including stroke detection on imaging that was "undetectable even 15 years ago."

He set the political frame: President Trump's AI executive orders, followed by Secretary Kennedy's direction that HHS take an "AI-forward" approach. The departmental goal: improve access, affordability, and impact of health care through technology.

Keane then offered the webinar's organizing metaphor, drawn from the Make America Healthy Again (MAHA) agenda. Health care has two traditional parts—lifestyle (the left circle: get active, eat real food, reduce screen time; Kennedy's getactive.gov and realfood.gov; the "85% of health that happens outside the medical system") and care infrastructure (the right circle). To these he added a third lever: artificial intelligence. He enumerated AI use cases now in play: stroke and imaging detection; ambient clinical note-taking (scribes); low-volume administrative tasks; clinical-trial matching and enrollment; drug discovery; syndromic surveillance for public health; and patient empowerment/agency. The intersection of clinical, administrative, and patient-agency use cases, he argued, yields "healthier, empowered Americans" inside a more accessible, more affordable system.

He closed by naming the "One HHS" coordination theme and noting that, while the RFI addressed regulation, reimbursement, and R&D, the webinar would cover regulation and R&D only.

2. The Cost-and-Chronic-Disease Case — Dr. Mark Atalla

Deputy National Coordinator (ONC)

Atalla supplied the "why": cost and quality. Drawing on CMS Office of the Actuary National Health Expenditure data (2009–2024), he cited:

  • A U.S. population of 341 million in 2024 and total national health expenditures of $5.3 trillion, spanning provider salaries, medications, labs, devices, administrative costs, and technology.
  • Per-capita health spending inflating over 7% annually in 2023 and 2024, above the historical trend.
  • Medicare at roughly 69 million beneficiaries and ~$1.2 trillion in 2025, projected by CMS to reach 78 million beneficiaries and $2.1 trillion by 2032.

The stated objective was to bring down premiums—commercial, individual, and Medicare Part B—over both near and long horizons.

On the long term, Atalla tied chronic-disease reversal to lifestyle (activity, real food, social connection), crediting pharmacotherapy where it has worked (statins lowering high-cholesterol prevalence) while noting CDC data showing rising chronic kidney disease and diabetes.

His diabetes deep-dive: roughly 115 million Americans with prediabetes and ~40 million with type 2 diabetes, against a system built to manage hemoglobin A1c. He observed that five of the top ten Part D drugs in 2023 were diabetes treatments, and that wait times to assemble the standard cardiology/endocrinology/ophthalmology team run 30–60+ days even in New York City. Trend data, he said, show fewer patients at goal A1c across age and sex, implying greater downstream risk of heart disease, kidney disease, and neuropathy.

On the short term, he pointed to the prediabetes population (which he later rounded to ~150 million) and a study comparing an AI-powered prediabetes coach against a human coach, reporting that the AI-led program was "not inferior" to the human-led program.

He framed the trajectory with Amara's Law (Roy Amara: we overestimate a technology's effect in the short run and underestimate it in the long run), distinguishing the possible, the probable, and the preferable—and argued that beyond today's administrative wins (scribes, revenue-cycle management), the real prize is medium-to-long-term AI augmentation of care in a "deflationary" direction.

3. AI Policy Timeline and Asks — Arman Sharma

HHS Deputy Chief AI Officer

Sharma narrated the administration's AI action timeline: a first national AI strategy in 2019 directing every federal agency, HHS included, to treat AI as a national priority; the removal of predecessor "barriers" on the third day of the second term and an order for a plan for "American AI dominance"; and, in summer 2025, America's AI Action Plan, built on three pillars—accelerating innovation, building infrastructure, and leading in AI diplomacy and security. He then walked the mission-specific sequence: a September executive order aimed at pediatric cancer (building on NCI's Childhood Cancer Data Initiative); the November "Genesis Mission," likened to the Manhattan Project, marshaling large scientific datasets and compute to accelerate AI-driven discovery; and a December move to clear the "patchwork" of conflicting state AI laws.

His core argument: the distinguishing feature of this push is adoption—it is not enough to build or power AI tools; they must be used. HHS, he said, operates three levers—regulation, reimbursement, and R&D—and the December RFI began the department's engagement across all three.

From the RFI he reported three needs heard (at a high level, not as a comment analysis):

  1. Coordination — the "One HHS" rationale; stakeholders deserve clarity on where, how, and when to engage the department.
  2. Adoption support — implementation and governance; closing the "implementation gap" and the workflow optimization required to bring AI into settings from small clinics to large academic medical centers.
  3. Evaluation and benchmarking — pushing the technical frontier to quantify what "good" means clinically, so patients and providers can judge tools and contexts.

Trust, he concluded, is the precondition for responsible and effective adoption. He closed with a pointed acknowledgment that data liquidity underpins clinical AI, crediting ONC and Keane for liberating patient data and for coordinating the RFI, external engagement, and AI efforts across the department.

4. Agentic AI for Cardiology — Dr. Haider Warraich

ARPA-H, ADVOCATE Program

Warraich—a cardiologist, ARPA-H program manager, and former FDA senior advisor for chronic disease—introduced ADVOCATE (Agentic AI-Enabled Cardiovascular Care Transformation). He situated ARPA-H as a DARPA-modeled agency built to take "ambitious swings" in health, and argued that although AI may be the defining technology of the age, patients don't yet see it affecting their care—held back by regulatory uncertainty and the risk of building tools that do "everything a clinician could do over the phone" while remaining safe, effective, and trusted within a regulated environment.

He chose cardiovascular disease—the leading cause of death and disability—because access is so thin: research from his team found that roughly 46% of U.S. counties have no cardiologist, and even where access exists, quality often falls short. The ADVOCATE vision is an agent that handles the full range of clinician-over-the-phone functions: administrative (scheduling, diet and lifestyle counseling) and clinical (diagnosis, triage, and high-risk actions including changing or prescribing medications). The initial focus is heart failure, but the intent is a template extensible to any chronic disease.

ADVOCATE comprises three technical areas, each an independent funding stream:

  1. A patient-facing cardiovascular/heart-failure agent — one patients talk to and that "closes the loop" by taking action rather than merely referring out, integrated in real time with the electronic health record and with wearables (rendered "variables" in the transcript), creating a proactive surveillance layer.
  2. A supervisory layer — to optimize human oversight of these tools in the post-market setting, addressing the safety concerns that gate adoption.
  3. Health-system co-development — recruiting health systems to resource and co-build the technology and deploy it in their clinical settings via randomized clinical trials.

The ambition, in his words, is to build the entire stack needed to make clinical agentic AI real for every American.

5. ACL Aging and Disability Priorities — Mary Lazare

Senior Official Performing the Duties of the ACL Administrator (Administration for Community Living)

Lazare framed ACL's mission around people challenged to remain in—or return to—community living: primarily older adults and people with disabilities. She walked ACL's AI-relevant priorities:

  • Caregiving — ACL's number-one priority (detailed by Cronin below).
  • Whole-person health — AI plus wearables, and home modification (safety and convenience features—locking doors, lighting, scheduling) to let people remain at home.
  • Employment — only about a third of people with disabilities are employed; AI to match skill sets to employers offering reasonable accommodations. She noted older adults are the fastest-growing segment of the homeless population and increasingly seek employment.
  • Economic security — applying AI to navigation of Trump accounts, Trump RX, and ABLE accounts (the 529A-style vehicle for people with disabilities), and to Medicare enrollment assistance.
  • Protecting rights / preventing abuse — interagency work to intercept scam calls before they reach recipients, across phones and laptops.
  • Connecting people to resources — AI matching of call-in needs to available resources nationwide.

On scale: 58 million people are over 65; she used age 60 as her threshold because Older Americans Act eligibility begins there, and noted 12,000 Americans turn 60 every day. The caregiving challenge is 63 million family/unpaid caregivers, with direct-care and direct-support workers in chronic short supply and rising demand.

6. Caregiver AI Prize Challenges — Kelly Cronin

Deputy Administrator (ACL, Administrator for Community Living)

Cronin detailed ACL's caregiving AI work. The goals: improve quality of care at home, reduce caregiver burden, and strengthen caregiver infrastructure—via AI that is affordable, practical, safe, and scalable, deployed alongside sensors, wearables, remote monitoring, and robotics to extend a strained workforce without compromising quality or replacing human connection, and informed by caregivers' own voices.

The Caregiver AI Prize Challenge (announced months earlier, then in phase one) runs two tracks: one for family and professional caregivers doing the work in the home, and one for extending the caregiver workforce—provider organizations, home-care agencies, and HCBS providers—by expanding capacity and reach. The aim is an innovation-and-learning community that surfaces implementation best practices.

The challenge's judging principles: protect privacy, dignity, and choice; support human-in-the-loop accountability; support caregiver well-being and reduce burden; supplement (not replace) human connection; support person-centered care; promote safety, reliability, and transparency (including algorithmic reliability/transparency and performance); and ensure affordability and access.

She also described a second, more recently launched challenge targeting dually eligible Medicare-Medicaid beneficiaries (and those at risk of becoming Medicaid-eligible through spend-down) with complex needs—keeping them at home through comprehensive, person-centered services delivered by teams of community-care networks and health-care partners. It rewards AI that enables clinical-community integration and care coordination across the continuum, and innovative approaches for the near-dually-eligible with complex needs. The throughline is a deliberate, collaborative learning environment with grantees and partners.

7. FDA Themes: Dr. Rick Abramson — Director, Digital Health Center of Excellence (FDA, CDRH)

Abramson presented without slides, citing active policy development that limited specifics, and offered general directions drawn from public workshops, listening sessions, and comment. He laid out five themes shaping FDA's clinical-AI policy:

  1. Greater clarity — increasingly important as AI systems gain functional complexity, autonomy, and agency; FDA aims to be clear on what it regulates and how, what is required of sponsors, and its role pre- and post-market.
  2. Right-size, risk-proportionate regulation — accounting not only for the technology's evolution but for the evolution in how users interact with it ("we don't interact with AI tools the same way in 2026 that we did in 2023 or 2020").
  3. Lifecycle oversight — clinical AI often behaves differently post-market than pre-market; FDA remains committed to a total-product-lifecycle approach and to balancing pre- and post-market oversight, "reflected in the weeks and months to come."
  4. Preparing for the future — technology evolves on a scale of weeks to months, regulation on months to years; FDA expects to "release some ideas to the public for stakeholder comment in very short order."
  5. Policy coordination — across federal entities, between federal and state bodies, and with international regulators; he cited collaborations with ONC, CMS, NIH, ARPA-H, ACL, HHS leadership, standards bodies, professional societies, and international regulators.

He affirmed FDA's full commitment to the "One HHS" principles.

Close — Mark Atalla

Deputy ONC

Atalla recapped the through-line—ARPA-H on heart failure, ACL on caregiving and the looming long-term-care crisis, and FDA on clear, risk-appropriate, lifecycle-managed regulation—and previewed three deliverables "over the coming weeks and months": (1) a coordinated way to navigate HHS; (2) adoption support, governance frameworks, and responsible adoption with trust at the center; and (3) benchmarking and evaluation. He closed on the claim that predictive, generative, and assistive technologies will improve access and bring "deflationary powers" to costs—paired with a closing slide whose thrust was that an AI-enabled, accessible, affordable future is inevitable, and that the timing is up to us.

Observations for readers

  • The missing lever. The RFI's reimbursement axis—the one most directly governing how diagnostics, laboratory testing, and novel clinical AI actually get paid—was named but not developed. 
    • Anyone reading the webinar for signals on coding, coverage, or payment policy will find the cost-growth framing (Atalla) and the "deflationary" rhetoric, but no payment mechanism.
  • Landscape over feedback. Despite being tied to a closed RFI comment period, the event presented departmental activity rather than what stakeholders said. The "three needs" framing (coordination, adoption support, evaluation/benchmarking) is the only gesture toward synthesized input.
  • Agentic AI is the headline R&D bet. ARPA-H's ADVOCATE—autonomous agents authorized to change and prescribe medications, with a dedicated post-market supervisory layer—is the most consequential single program surfaced, and the one most likely to force FDA's lifecycle and autonomy questions into the open.

Editorial note

This record is built from an auto-generated (otter.ai) transcript of the webinar, with speaker names and titles corrected against official HHS, ARPA-H, ACL, and FDA sources, and obvious transcription artifacts repaired.