The US government's health research agency is moving to create something that does not yet exist: an artificial intelligence agent, authorized by the Food and Drug Administration, that can deliver specialty medical care 24 hours a day. The Advanced Research Projects Agency for Health (ARPA-H) has launched a program called ADVOCATE, short for Agentic AI-Enabled Cardiovascular Care Transformation, aiming to develop the first FDA-authorized agentic AI technology that can provide round-the-clock specialty care.
For the tens of millions of people living with heart disease, the leading cause of death in the United States, the pitch is access: a care companion that is available at any hour, not only during a clinic appointment that may be weeks away. For the AI field, ADVOCATE is a test of a much harder problem than chatbots have faced, because an agent that takes actions in a patient's care carries a level of risk that an information tool does not.
What the program would build
ADVOCATE plans to select teams of innovators, reportedly within about six months, to create a patient-facing AI agent focused on cardiovascular disease. The agency's stated goal is for the agents to supplement clinicians rather than replace them, by handling tasks such as scheduling appointments, adjusting medications, and supporting diet and exercise, while connecting to the electronic health record and integrating with wearable devices so the agent has continuous data and is available to patients 24/7.
That description is what makes this notable. Most medical AI cleared to date is a tool a clinician uses, an imaging algorithm that flags a suspicious scan, or a documentation assistant. ADVOCATE describes an agent that acts: it would not just suggest a medication change but carry parts of the care plan forward on its own. That is the distinction between "agentic" AI and an advisory chatbot, and it is precisely why the regulatory path is treated as a multi-year effort, reported at roughly three years, with a "down select" after the first year that narrows the field to the most promising approaches.
Why an always-on agent is appealing in cardiology
Cardiovascular disease is both extremely common and heavily dependent on continuous management: medication titration, blood-pressure tracking, lifestyle adjustment and timely follow-up. Specialist cardiologists are a limited resource, and the gaps between visits are where patients often fall out of good control. An agent that monitors wearable data, nudges patients, and adjusts routine elements of a care plan in real time could, in principle, catch problems earlier and extend a specialist's reach to far more patients than a clinic can see.
The program also fits a broader regulatory shift. The FDA has been moving from a one-time approval model toward continuous, adaptive oversight for AI medical tools, recognizing that models change after they are deployed. The agency has even begun deploying agentic AI in its own operations, a sign that regulators are wrestling with the same technology they will be asked to authorize.
The risks that come with an agent that acts
The same feature that makes ADVOCATE ambitious, an agent that takes actions, is the source of its hardest problems. An AI that adjusts medications is operating in territory where errors can cause direct physical harm, which raises questions the technology has largely avoided so far.
Liability is unresolved: if an agent's medication adjustment contributes to harm, responsibility among the developer, the supervising clinician and the health system is not clearly settled. Bias is a documented concern in clinical AI, where models trained on unrepresentative data can perform worse for some patient groups and, in the worst case, amplify existing disparities in care. Automation bias is another: clinicians and patients may over-trust an agent's recommendations precisely because it is available and confident. And the data foundation, an agent wired into the electronic health record and continuous wearable streams, concentrates highly sensitive health information and broadens the privacy and security surface.
None of these is a reason the effort should not happen; they are the reasons it is structured as a long, staged program with FDA authorization as the gate rather than a fast product launch. The three-year timeline is not bureaucratic drag so much as an acknowledgment that an agent acting on patients has to clear a far higher bar than software that only informs.
What it means for patients now
In the near term, nothing changes at the bedside; ADVOCATE is a development program, not a product patients can use today. Its value to readers is as a signal of where regulated medical AI is heading: from tools that assist clinicians toward agents that carry parts of care, with the FDA building the oversight model in parallel. If it succeeds, the first beneficiaries would be cardiovascular patients who struggle to maintain continuous specialist contact. If it stumbles, the lessons will shape how every future clinical AI agent is judged.
Bottom line
ARPA-H's ADVOCATE program aims to build the first FDA-authorized agentic AI for 24/7 cardiovascular care, an agent that would adjust medications, schedule care and monitor wearables, on a roughly three-year regulatory path. The promise is wider access to continuous heart care; the challenge is that an AI which acts on patients raises questions of liability, bias and safety that advisory AI never had to answer. The program's staged, FDA-gated design is the clearest sign of how high those stakes are.
This article describes a federal research program; it is not medical advice. Patients should not change any medication or treatment based on it and should consult their clinician.
Frequently Asked Questions
What is ADVOCATE? A new ARPA-H program, Agentic AI-Enabled Cardiovascular Care Transformation, aiming to develop the first FDA-authorized AI agent that can provide 24/7 cardiovascular care, supplementing clinicians.
What would the AI agent do? According to the program, schedule appointments, adjust medications, support diet and exercise, connect to the electronic health record and integrate with wearables to monitor patients continuously.
How is this different from a medical chatbot? A chatbot gives information. An agentic AI takes actions in a patient's care, which carries far higher risk and is why the effort is on a multi-year, FDA-authorized path rather than a quick launch.
When will it be available? There is no consumer product yet. The program plans to select teams and follows a roughly three-year approval timeline, with an early "down select" to narrow approaches after the first year.
What are the main concerns? Liability for an agent's actions, bias in clinical AI, over-reliance on automation, and the privacy and security of health and wearable data the agent would access.
ⓒ 2026 TECHTIMES.com All rights reserved. Do not reproduce without permission.
