ARPA-H, the federal government's advanced research agency for health technology (modeled after the Pentagon's DARPA), launched a program called ADVOCATE in January 2026 to fund the development of clinical AI agents that can autonomously manage cardiovascular disease. The full proposal deadline closed Tuesday at 5 p.m. ET. The program has received nearly 300 solution summaries, making it one of the most oversubscribed in ARPA-H's short history. If even one team succeeds, it would mark the first time the FDA has authorized a clinical AI agent — not a predictive algorithm, but a system that acts.
The distinction matters. The FDA has already authorized more than 1,000 AI-enabled medical devices. Every single one is predictive: it analyzes data and outputs a diagnosis, a risk score, a flag for a human to act on. ADVOCATE is explicitly seeking something different — an AI agent that can write or modify prescriptions autonomously, within a defined scope, without waiting for a physician to sign off. That is a different regulatory category, and no device has crossed into it yet.
"Heart failure is the prototypical condition where we know what works, we know what doesn't work, and the gap between the two is enormous," Haider Warraich, program manager at ARPA-H and a practicing cardiologist at Boston Medical Center, told Stat News in January. Warraich also served as senior advisor for chronic disease to the FDA commissioner between 2023 and 2024. Only 2.5 percent of American heart failure patients currently receive recommended guideline-directed care, he said. The rest fall through gaps in access, adherence, and follow-up that a properly designed AI system could close.
The geographic dimension is part of the problem ADVOCATE is trying to solve. Nearly half of all U.S. counties have no cardiologist, according to Stat News reporting. An AI agent that can manage medication titrations, monitor adherence through connected devices, and adjust prescriptions within guardrails could bring specialist-level care to counties without any. The program's pitch includes a potential $50 billion in annual savings if successful, per ARPA-H's own projections.
ADVOCATE is structured around three technical areas. The first is a patient-facing clinical AI agent that can independently write or modify in-scope prescriptions. The second is a supervisory agent that monitors the first for safety and drift. The third is a health systems integration layer to ensure the system works within existing hospital and clinic infrastructure. Warraich has noted that ARPA-H is piloting secure large language model tools to assist with the initial review of solution summaries, consistent with the program's broader vision of enabling clinical AI agents that can safely operate.
The architecture raises obvious safety questions that the program has not fully resolved. John Whyte, CEO of the American Medical Association, told Fierce Healthcare that using autonomous AI to deliver cardiovascular care without clear physician oversight "raises serious concerns about patient safety." The AMA's position is not a fringe view — it reflects the mainstream medical establishment's position on where autonomous AI prescribing should sit relative to physician authority. ADVOCATE's supervisory agent layer is designed to address exactly this concern, but whether it satisfies regulators is the central open question.
The program timeline is tight. ARPA-H plans to select winning teams in June 2026, with the goal of shepherding them through FDA authorization within two years of contract award. The full program, including FDA approval, is slated to last 39 months. For context: the FDA has never authorized a clinical agentic AI device. The fastest prior authorizations for AI-enabled devices have taken multiple years, and those were predictive systems with more established risk profiles.
The program attracted enough interest to nearly saturate ARPA-H's review capacity before the first technical area award has even been made. The proposal deadline was April 1, 2026 at 5 p.m. ET. Whether that interest reflects genuine clinical readiness or the availability of federal grant funding for AI agent research is a question the program's outcomes will eventually answer.
What ADVOCATE is building, if it works, would be a regulatory and clinical infrastructure test case for the entire class of agentic medical AI. It is not a product announcement. It is a bet that the FDA's existing authorization framework can accommodate a system that acts, rather than one that only recommends. That bet has not yet been cashed.
