The Advanced Research Projects Agency for Health on Wednesday committed $62.7 million to build artificial intelligence systems that could eventually prescribe drugs and order tests for heart failure patients without a specialist in the room, a bet that regulatory-grade automation can plug a clinical gap affecting 6.7 million Americans.
The money and the milestones
The first year carries $33.7 million across Atman Health, UpDoc, Tempus AI, Stanford, Duke, and Kaiser Permanente. The remaining $29 million sits in a renegotiable tranche, meaning the final figure could move up or down depending on whether the agency likes what it sees. ARPA-H left the door open for additional teams.
The regulatory hook
The program, called ADVOCATE, is not aiming for decision support. It is aiming for partial autonomy with FDA authorization, a distinction that matters because reimbursement pathways for autonomous AI do not really exist yet. If the devices clear the agency, the commercial question becomes who pays for a bot that adjusts diuretics.
The access argument
The clinical rationale is straightforward: most heart failure patients never see a cardiologist, and rural hospitals are not staffing up. The counterpoint is that the sickest patients, the ones driving readmissions, are also the ones least likely to be managed safely by an algorithm that has never been tested in a prospective trial. ARPA-H is funding the trial.
What to watch
Tempus AI is the only public company in the first cohort, and its shares will be the market's real-time proxy for whether investors believe the FDA will bless autonomous prescribing. The next inflection point is not a data readout but a regulatory meeting, whenever the first team sits down with the agency to define what "partially autonomous" actually means.