ARPA-H ADVOCATE: Inside the Federal Push to Build the First FDA-Authorized AI Doctor

Forty-six percent of U.S. counties do not have a single cardiologist. Heart disease kills more than 700,000 Americans every year — one person every 33 seconds — and the gap between the number of patients who need specialized cardiac care and the number of physicians available to deliver it is widening. The Advanced Research Projects Agency for Health thinks an AI agent might be the answer.

ARPA-H's ADVOCATE program — Agentic AI-Enabled Cardiovascular Care Transformation — is not another chatbot pilot or diagnostic screening tool. It is a direct attempt to build autonomous AI systems that can prescribe medications, adjust treatment plans, schedule appointments, and provide real-time clinical guidance to heart failure patients, all without a human physician in the loop for routine decisions. If it works, it would be the first agentic AI technology to receive FDA authorization for clinical use.

Full proposals are due April 1, 2026. The stakes, the technical requirements, and the competitive dynamics make this one of the most consequential health technology funding opportunities in years.

What ADVOCATE Actually Requires

The program is structured around three distinct technical areas, and ARPA-H is deliberately keeping them separate to prevent conflicts of interest.

Technical Area 1: The Patient-Facing AI Agent. This is the centerpiece. Teams must build an autonomous clinical agent capable of connecting to electronic health records, integrating data from wearable monitors, diagnosing conditions, writing and modifying prescriptions, providing dietary and physical therapy recommendations, and determining when to escalate to a human provider. The system must demonstrate non-inferiority compared to board-certified cardiologists — meaning it must perform at least as well as human specialists across measurable clinical outcomes.

That last requirement is the critical bar. ARPA-H is not asking for a decision-support tool that flags abnormalities for a doctor to review. They want a system that can independently manage a patient's cardiovascular care around the clock.

Technical Area 2: The Supervisory Agent. A separate, disease-agnostic AI system that monitors the patient-facing agent for safety, accuracy, and risk. ARPA-H has expressed a strong preference for open-source solutions in this track — a deliberate choice that would allow the supervisory architecture to be adapted for other clinical domains beyond cardiology. Think of it as an AI watchdog for AI doctors.

Technical Area 3: Clinical Integration. Health systems that will serve as real-world deployment sites for the technology. These partners must provide infrastructure for large-scale studies and generate clinical outcome evidence. Importantly, TA3 applicants cannot also apply for TA1 or TA2, ensuring that the organizations validating the technology are independent from those building it.

The FDA Pathway Nobody Has Walked

The most ambitious element of ADVOCATE is the explicit commitment to FDA authorization within roughly three years. Selected teams will move through two phases — Phase 1A (12 months) and Phase 1B (12 months, optional) for development and initial validation, followed by Phase 2 (15 months, optional) for clinical deployment and regulatory submission. The total performance period caps at 39 months.

No agentic AI system has ever received FDA authorization for autonomous clinical decision-making. The closest precedent is Aidoc's recent clearance of a foundation-model-powered system that triages 14 acute conditions on CT scans — a major milestone, but one that still operates within a diagnostic-assistance framework rather than autonomous patient management.

ADVOCATE wants to leap past that boundary entirely. The program's AI agents would not just detect problems — they would treat them, adjusting medications and care plans in real time based on continuous patient monitoring.

ARPA-H Director Alicia Jackson has framed ADVOCATE as a test case for how the federal government can accelerate regulatory pathways for health AI. Program Manager Haider Warraich, a cardiologist himself, describes the envisioned system as "a clinician-extender: an autonomous agent smart enough to understand patients' treatment needs." HHS Deputy Secretary Jim O'Neill is directly involved, signaling that this has institutional backing well above the typical program-manager level.

Who Should Apply — And Who Probably Shouldn't

ADVOCATE is funded through Other Transaction Agreements, not traditional grants or contracts. This mechanism gives ARPA-H flexibility to negotiate terms, milestone payments, and IP arrangements directly with performers, but it also means the process is less standardized than a typical NIH R01 or SBIR Phase II.

Eligible applicants include U.S.-based startups, universities, nonprofits, and non-federal research organizations. Federal employees, federally funded research and development centers (FFRDCs), and entities with foreign ownership concerns are excluded.

The practical reality is that competitive proposals will need to combine deep clinical AI expertise with existing EHR integration capabilities and established relationships with health systems for TA3 deployment. A university lab with a promising algorithm but no pathway to clinical deployment will struggle. A health-tech startup with a working EHR integration layer but no cardiology AI capabilities will need to team up.

ARPA-H has explicitly encouraged teaming, and the mandatory Solution Summary that was due February 27 was partly designed to facilitate matchmaking among complementary organizations.

The $55 Billion Question

ARPA-H estimates that successful deployment of ADVOCATE-type technology could save $55 billion annually in cardiovascular healthcare costs. That figure reflects the enormous inefficiency of the current system: patients with heart failure cycle between emergency rooms, specialist offices, and home care with minimal coordination, often going weeks between meaningful clinical interactions.

An AI agent that provides continuous monitoring and real-time treatment adjustments could, in theory, catch decompensation events before they require hospitalization, optimize medication regimens faster than quarterly office visits allow, and extend specialty-level cardiac care to the 46 percent of counties that currently have no cardiologist at all.

The potential is real, but so are the risks. Autonomous prescription authority for an AI system raises questions about liability, malpractice, and patient safety that the healthcare system has never had to answer. If an ADVOCATE agent prescribes a beta-blocker dose adjustment that causes an adverse event, who is responsible — the AI developer, the deploying health system, or the supervisory agent that approved the decision?

These questions will need answers before any FDA authorization panel, and the teams that address them most convincingly in their proposals will have a significant advantage.

The Broader Signal for Health AI Funding

ADVOCATE does not exist in isolation. It reflects a coordinated federal strategy to push health AI beyond diagnostics into autonomous clinical operations. The FDA's recent clearance of Aidoc's multi-condition foundation model for CT triage, CMS's Health Tech Ecosystem initiative promoting standardized data-sharing frameworks, and ARPA-H's own portfolio of 23 programs across 150 projects are all pieces of the same puzzle.

For grant seekers in health technology, the message is clear: the federal government is willing to fund AI systems that do things, not just flag things. The diagnostic-only paradigm that defined the first wave of health AI is giving way to a more ambitious vision where AI agents operate as semi-autonomous clinical actors.

ARPA-H's FY2026 budget sits at $945 million — down from $1.5 billion in FY2025, though a coalition of 77 organizations is pushing Congress for $1.7 billion. Even at the reduced level, the agency's five focus areas for 2026 (chronic disease, rural access, proactive health, healthcare security, and frontier health technologies) all create openings for AI-forward proposals.

As Granted News reported, ADVOCATE represents a landmark moment for agentic AI in healthcare. But the real significance may be less about cardiology and more about precedent. If the program produces an FDA-authorized autonomous clinical agent — even a narrow one focused on heart failure management — it creates a regulatory and technical template that every other medical specialty will want to replicate.

What to Do Before April 1

For teams considering a full proposal submission, the clock is running. Here is what the next 30 days should look like:

For AI developers (TA1): Your architecture must be agentic, not rules-based. ARPA-H has been explicit about this distinction. If your system operates on decision trees or fixed protocols, it does not qualify. You need a system that reasons about patient state, generates treatment plans, and adapts autonomously. EHR integration capability is non-negotiable — proposals without a credible technical path to Epic, Cerner, or comparable systems will not advance.

For safety and monitoring teams (TA2): The open-source preference is significant. Building a proprietary supervisory agent that locks in a single vendor will work against you. ARPA-H wants a monitoring framework that can be adapted across disease domains, and they want the broader research community to be able to build on it.

For health systems (TA3): You need to demonstrate diverse patient populations, existing digital health infrastructure, and a willingness to run large-scale clinical studies. Rural health systems have a particular advantage here — ARPA-H's emphasis on reaching underserved counties means that proposals from academic medical centers alone, without rural deployment partners, may fall short.

The full solicitation is posted as ARPA-H-SOL-26-142 on SAM.gov. Given the 39-month performance period and the explicit FDA authorization target, this is not a research grant — it is a product development contract with regulatory milestones. Teams that approach it accordingly will write the strongest proposals.

Tools like Granted can help you identify complementary funding opportunities and build submission-ready proposals while the window is still open.