For years, AI has mostly played a supporting role in clinical care — flagging a suspicious scan, drafting a note, summarizing a chart. Federal health officials are now reaching for something far more autonomous: AI agents that can take clinical action on their own, including diagnosing patients and adjusting their medications.
That ambition surfaced June 25, when leaders from across HHS gathered online to share what they had heard from a December request for information on accelerating the adoption of AI in clinical care, which drew more than 7,000 public comments. The event was organized around a coordinated, cross-agency approach the department calls “one HHS,” building on an AI strategy it released late last year.
Thomas Keane, MD, national coordinator for health IT, opened by recounting how AI now helps physicians spot strokes on imaging in seconds — work that was effectively impossible two decades ago.
“Technology has really transformed clinical care, and a big part of that technology is AI,” he said.
But the tools Dr. Keane described still assist a physician. The most concrete vision of AI acting on its own came from the Advanced Research Projects Agency for Health.
Haider Warraich, MD, a cardiologist and program manager at ARPA-H, is overseeing a program called ADVOCATE — short for Agentic AI-Enabled Cardiovascular Care Transformation — that aims to build the first FDA-authorized agentic AI capable of delivering around-the-clock cardiovascular care.
The technology would be patient-facing and integrated in real time with the EHR and data from wearable devices, and it would act without routing every decision back to a clinician. Beyond scheduling appointments and offering diet and lifestyle guidance, Dr. Warraich said, the agents would diagnose patients, perform triage and handle higher-risk tasks such as changing or prescribing medications.
“This technology will be able to close the loop, instead of directing them to a clinician,” Dr. Warraich said. “It’ll be able to take the actions that the patient needs.”
He described the goal as building software that could “do everything a clinician can over the phone.” The rationale, he said, is access: Nearly half of U.S. counties do not have a single cardiologist.
Although ADVOCATE is focused on patients with heart failure and those who have had heart attacks, Dr. Warraich said the broader aim is a model others can follow.
“We are building a template that would be applicable to any patient with chronic disease,” he said.
He framed the work in sweeping terms, describing it as building “the entire stack that will be needed to make clinical agentic AI a reality for every American.”
For health systems, the harder questions begin after deployment, including who is accountable when an autonomous agent gets something wrong, a scenario some systems are already grappling with.
“Clinicians will have a lot of concern about technology that is so powerful,” Dr. Warraich said.
ADVOCATE is funding a second, “supervisory” agent meant to monitor clinical AI agents once they are deployed and flag when human oversight is needed.
The federal government is trying to build the regulatory scaffolding at the same time. Rick Abramson, MD, director of the FDA’s Digital Health Center of Excellence, said the agency is preparing for AI tools that operate with “greater autonomy and greater agency” and is working toward “regulation that is proportionate to risk.” Oversight cannot stop at approval, he said, because clinical AI tools can “behave very differently in the post-market setting from how they behaved pre-market.”
Dr. Abramson said the FDA expects to “release some ideas to the public for stakeholder comment in very short order,” though he declined to detail specific policy directions while they remain under development.
None of it is available at the bedside yet. ADVOCATE is a research and development effort, not a product patients can use today; ARPA-H has described a multiyear program with early rounds of winnowing before any technology is tested in clinical trials.
For now, the message to health systems is that autonomous clinical AI is coming and that Washington intends to shape how. The unresolved question — one the agencies acknowledged they are still working out together — is how to let an AI act on a patient’s care without losing the trust of the clinicians it is meant to extend.
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
