News
ARPA-H selects teams for ADVOCATE, a 39-month program to build FDA-authorized agentic AI for cardiovascular care
ARPA-H launched ADVOCATE (Agentic AI-EnableD CardioVascular CAre TransfOrmation) in January 2026 and selected performer teams in June 2026 following a competitive solicitation. The 39-month, two-phase program targets patients with heart failure and prior myocardial infarction, aiming to create a first-of-its-kind FDA-authorized clinical agentic AI system that autonomously adjusts appointments, medications, diet, and exercise recommendations around the clock. A supervisory agent will monitor clinical AI agents for safety. ARPA-H cited that 46% of U.S. counties lack a cardiologist as motivation.
- ARPA-H
- cardiovascular
- agents
- agentic-AI
- FDA
- federal
ADVOCATE is arguably the most consequential federal AI-in-healthcare program announced in the current funding cycle, for two reasons that have nothing to do with the dollar amounts. First, it is explicitly aiming for FDA authorization for a clinical agentic AI system — not a diagnostic tool, not a triage flag, but an agent that autonomously adjusts medications and appointments for patients with heart failure. That is a regulatory first. No agentic AI system has an FDA clearance or approval pathway precedent. Second, it is structured with a 39-month two-phase timeline and competitive down-selections, which means ARPA-H is treating this as an engineering program rather than a research grant — the expectation is working systems, not publications.
The 46% of U.S. counties lacking a cardiologist statistic is the motivating constraint ARPA-H keeps citing, and it frames the access-equity case for autonomous agentic AI more clearly than most clinical AI pitches. Heart failure is a condition where the monitoring burden is high (daily weight checks, symptom tracking, medication titration, diet management) and where the gap between what evidence-based protocols recommend and what actually happens in rural primary care without specialist access is large and quantifiable in avoidable hospitalizations. An agentic AI that can close that gap even partially has a different cost-effectiveness argument than a diagnostic AI that improves detection rates.
The supervisory agent architecture — a separate AI that monitors the clinical AI agents for safe and effective recommendations — is interesting from a safety engineering standpoint and reflects the evolving consensus that multi-agent safety in healthcare requires explicit oversight layers rather than monolithic single-agent trust. This mirrors the architectural patterns emerging in commercial health AI platforms as well (see Bunkerhill Health’s Carebricks), where agent governance and auditability are becoming first-class design requirements.
The ADVOCATE timeline puts FDA authorization attempts in Phase 2 (starting around month 25). How the FDA handles the first 510(k) or De Novo submission for an autonomous medication-titrating agentic AI will set a precedent that shapes the entire category. The agency’s recent PCCP guidance — enabling predefined post-market model updates without new submissions — addresses a different problem (model drift and improvement), not the initial authorization bar for autonomous clinical action.
Related coverage: FDA & devices topic.
Primary source: Read the full original on ARPA-H ↗