The US Government Wants AI Doctors. ARPA-H Is Building Them With a 3-Year FDA Fast Track.

AI That Can Practice Medicine, Autonomously

The Advanced Research Projects Agency for Health (ARPA-H) just laid out one of the most ambitious healthcare AI projects ever attempted by the U.S. government. The goal: build autonomous AI agents that can provide 24/7 cardiovascular care to patients, guide treatment decisions, and interact with patients directly. Not as a suggestion engine for doctors. Not as a chatbot that says "consult your physician." As a clinical agent that can autonomously guide cardiovascular disease care.

The program is called ADVOCATE, short for Agentic AI-Enabled Cardiovascular Care Transformation, and the timeline is aggressive. ARPA-H will select development teams by June 2026, run a competitive down-select process after 12 months, and aim to have the entire project, including FDA authorization, completed within 39 months, just over three years. If it works, it would be the first FDA-authorized agentic AI technology operating in a high-risk clinical setting.

This is happening at the same time the HIMSS 2026 conference in Las Vegas is consumed by a single question: AI tools are improving faster than regulators can evaluate them, so who decides when they're safe enough to let loose in hospitals?

What ADVOCATE Actually Does

The ADVOCATE program envisions a patient-facing AI agent for cardiovascular disease. Think of it as an AI cardiologist that can monitor a patient's condition around the clock, adjust treatment recommendations based on real-time data, flag emergencies before they become critical, and communicate directly with patients about their care plan.

Cardiovascular disease kills more Americans than any other condition, roughly 695,000 per year. Access to specialized cardiology care is unevenly distributed, with rural areas and underserved communities often lacking the specialists needed for timely intervention. An AI agent that can provide continuous, expert-level monitoring would be transformative for these populations.

But ARPA-H isn't just building the AI agent itself. The agency is also funding the development of a supervisory agent, essentially an AI that watches the clinical AI to ensure it continues to perform safely and effectively. This layered approach reflects a growing understanding in the field that autonomous AI systems need their own oversight mechanisms, particularly when the stakes are life and death.

The program structure is designed to maximize competition. Multiple teams will be selected in June, given resources to develop their approaches, and then narrowed down based on performance. This is the same DARPA-style "shoot-off" model that has produced breakthroughs in defense technology, now applied to healthcare.

The Trump Administration's Regulatory Bet

The political context matters enormously. The Trump administration has taken a deliberately hands-off approach to AI regulation, choosing to limit rules that could slow adoption rather than build comprehensive oversight frameworks. The Department of Health and Human Services issued a public request for feedback on how to "accelerate the adoption and use of AI as part of clinical care," with the emphasis clearly on speed rather than caution.

This philosophy creates both opportunity and risk. The opportunity is that the 39-month timeline for ADVOCATE would be impossible under a more restrictive regulatory environment. Traditional FDA medical device approval can take years or even a decade, and the iterative nature of AI systems (which improve through updates and retraining) doesn't fit neatly into the existing framework of one-time pre-market review.

The risk is that moving fast with autonomous clinical AI means accepting a higher tolerance for error during the development phase. The FDA's existing framework requires developers to notify the agency about update plans, and AI tools that can potentially improve themselves create a regulatory paradox: the product that gets approved today may be fundamentally different from the product operating six months later.

ARPA-H's approach of building a supervisory AI agent alongside the clinical agent is a creative solution to this problem. Instead of relying solely on regulatory checkpoints, the system is designed with continuous internal monitoring. Whether that's sufficient for a technology making life-or-death decisions remains an open and urgent question.

The HIMSS Reality Check

At the HIMSS 2026 conference this week, the tension between innovation and oversight was on full display. Experts acknowledged that AI tools are improving so rapidly that traditional regulatory timelines are becoming obsolete. But they also warned that the absence of clear guidelines is creating a patchwork of adoption practices across hospitals and health systems.

Some organizations are deploying AI aggressively, using it for diagnostic imaging, clinical documentation, treatment recommendations, and even patient communication. Others are holding back, waiting for regulatory clarity that may never come under the current administration. The result is a fragmented landscape where the quality and safety of AI-assisted care varies dramatically depending on where you live and which health system you use.

The healthcare industry has learned this lesson before. Electronic health records were pushed into widespread adoption through federal incentives without adequate standardization, creating interoperability problems that persist to this day. The fear is that AI adoption without clear safety standards could produce a similar mess, except with higher stakes because the technology is making clinical decisions rather than just storing data.

The Global Context

The U.S. isn't acting in isolation. The EU's AI Act, which took effect in stages through 2025, classifies medical AI as "high-risk" and imposes extensive pre-market requirements including conformity assessments, human oversight provisions, and ongoing post-market surveillance. China has its own regulatory framework through the NMPA (National Medical Products Administration) that requires extensive clinical trial data before AI medical devices can be marketed.

The contrast is stark. Europe and Asia are building frameworks that prioritize safety verification before deployment. The U.S. under the current administration is prioritizing deployment speed with post-market surveillance as the primary safety mechanism. ADVOCATE represents the most extreme version of this approach: not just approving existing AI tools faster, but actively building autonomous AI agents under an accelerated timeline.

The question of which approach produces better health outcomes will take years to answer. But the competitive dynamics are clear. If ARPA-H's program succeeds, the U.S. will have the first FDA-authorized autonomous clinical AI, creating a template that other countries could adopt. If it fails, particularly if there are patient safety incidents, it could set back clinical AI adoption globally and validate the more cautious approaches taken elsewhere.

Why This Matters Beyond Healthcare

ADVOCATE isn't just a healthcare project. It's a test case for how governments approach autonomous AI in any high-stakes domain. If you can build an AI agent that autonomously guides cardiovascular care and get it through FDA approval in 39 months, the same framework could apply to AI agents in air traffic control, nuclear plant monitoring, financial system oversight, or any other domain where autonomous decisions carry significant risk.

The supervisory agent concept is particularly interesting. Rather than relying on human oversight for every AI decision (which becomes impractical as systems scale), ARPA-H is pioneering a model where AI systems monitor other AI systems. This raises its own questions about accountability, transparency, and failure modes, but it's also the most realistic path toward deploying autonomous AI at scale.

The teams will be selected in June 2026. The first down-select happens in mid-2027. Full FDA authorization is targeted for late 2029. Along the way, every success and failure will be scrutinized by regulators, researchers, and companies around the world, because ADVOCATE isn't just building a cardiovascular AI agent. It's building the playbook for autonomous AI in critical systems.

What to Watch

Watch for the team selection announcement in June. The organizations chosen will signal whether ARPA-H is betting on traditional medical device companies, AI-native startups, academic medical centers, or some combination. Watch the FDA's response as the program progresses; the agency hasn't publicly committed to the 39-month timeline, and pushback from career staff could slow the process. And watch the data coming out of HIMSS and other healthcare conferences, where the debate between speed and safety will only intensify as ADVOCATE moves from concept to reality.

The first AI doctor may not arrive in your hospital anytime soon. But the project to build one just got a three-year deadline and a government backing that makes it very, very real.