By Elise Reuter
Published September 11, 2026
In a landmark move to reshape the landscape of chronic disease management, the Advanced Research Projects Agency for Health (ARPA-H)—the high-stakes innovation arm of the U.S. Department of Health and Human Services—has officially launched the ADVOCATE program. This initiative aims to deploy "agentic AI" systems capable of functioning as autonomous members of a clinical care team, specifically designed to support heart failure patients during the critical intervals between physician visits.
By enlisting industry heavyweights including Tempus AI and Updoc, alongside premier research institutions like Stanford and Duke University, the agency is signaling a paradigm shift: moving from passive, reactive healthcare to a proactive, AI-driven model of continuous monitoring and intervention.
The Core Objective: Moving Beyond Traditional AI
At its heart, the ADVOCATE program seeks to redefine the capabilities of medical software. Unlike traditional AI, which often operates as a diagnostic support tool—analyzing static data to provide a risk score—"agentic AI" possesses the capability to perceive environmental changes, make autonomous decisions, and execute actions.
According to Dr. Haider Warriach, a cardiologist and the program manager for ADVOCATE, the vision is to create a digital surrogate for the clinical team. "The hope is to develop an agent that can provide certain care autonomously and engage a patient’s clinical team when needed," Warriach stated. By automating routine symptom checks, medication adherence tracking, and basic clinical triaging, these agents aim to reduce the massive burden on the U.S. healthcare system, particularly in regions where specialist access is non-existent.
The urgency is underscored by a sobering statistic: nearly 50% of U.S. counties lack a single practicing cardiologist. For the millions of Americans suffering from heart failure, this "care desert" phenomenon often leads to delayed treatment, preventable hospital readmissions, and a lower quality of life.
A Multi-Tiered Strategic Framework
The ADVOCATE program is structured into three distinct, specialized teams, each tasked with a specific pillar of the technology’s lifecycle: development, safety, and real-world deployment.
Team 1: The Patient-Facing Frontier
Led by private-sector innovators including Tempus AI, Updoc, and Atman Health, this team is responsible for the "front end" of the experience. They must develop an interface that interacts directly with patients, monitors their health status, and escalates concerns to human clinicians when specific thresholds are met.
Crucially, this group is operating under an aggressive two-year timeline. By the end of this period, they are mandated to submit a formal package to the U.S. Food and Drug Administration (FDA) for regulatory authorization. This represents a significant acceleration of the standard med-tech development lifecycle.
Team 2: The Supervisory Guardrails
Recognizing the risks inherent in autonomous AI, a team led by Stanford University has been tasked with building a "supervisory AI system." This system will serve as a watchdog for the patient-facing agents. Its primary goal is to ensure that the agents remain within "clinical bounds." This includes identifying "out-of-distribution" behavior—cases where the AI encounters data significantly different from its training set—and blocking unsafe recommendations before they reach the patient. This layer of oversight is intended to be "disease agnostic," potentially serving as a universal safety architecture for future medical AI applications.
Team 3: Real-World Deployment and Integration
The final pillar, led by Duke University and Kaiser Permanente, focuses on the messy, complex reality of clinical workflows. Duke is tasked with testing the technology across five disparate health systems, integrating the software into both Epic and Cerner/Oracle electronic health record (EHR) environments. Meanwhile, Kaiser Permanente will deploy the agents across 21 medical centers and over 260 clinics. This work is focused specifically on rural and underserved communities, ensuring that the technology performs effectively in the field, not just in a controlled laboratory setting.
Chronology of the ADVOCATE Launch
The rollout of the ADVOCATE program follows months of intensive vetting and interdisciplinary planning:
- Q1-Q2 2026: ARPA-H conducted a series of "listening sessions" with digital health stakeholders, focusing on the regulatory challenges of generative AI in clinical settings.
- August 2026: Final selection of the three-team consortium was completed, emphasizing a balance between commercial agility and academic rigor.
- September 11, 2026: Official public launch of the ADVOCATE initiative.
- Upcoming 24 Months: The primary development phase for Team 1, leading to an anticipated FDA submission.
- 2028 and Beyond: Phased rollout into rural health systems and large-scale longitudinal performance assessment overseen by Johns Hopkins University’s Applied Physics Laboratory.
Supporting Data and Technical Oversight
The integrity of the program is bolstered by the involvement of the Johns Hopkins University Applied Physics Laboratory (APL), which will serve as an independent auditor. APL will assess technical performance, latency in decision-making, and clinical outcomes.
Furthermore, the participating companies are mandated to work with ARPA-H to establish a "common language" for the industry. This includes:
- Shared Evaluation Standards: Defining what "success" looks like for an AI agent in terms of clinical accuracy.
- Interoperability Requirements: Ensuring that agents can "speak" to different EHR systems seamlessly.
- Reimbursement Pathways: Collaborating with CMS and private insurers to determine how physicians will be compensated for the time they spend reviewing AI-generated clinical escalations.
Official Perspectives and Regulatory Implications
The FDA, which has been increasingly vocal about the need for a new regulatory framework for generative AI, views the ADVOCATE program as a laboratory for future policy.
Rick Abramson, director of the FDA’s Digital Health Center of Excellence, noted the significance of this collaboration: "These kinds of efforts will be critical as we think about continuous ongoing monitoring of autonomous clinical AI for both safety and effectiveness."
The FDA’s interest in generative AI—which can create content and reach conclusions that are not strictly deterministic—poses a regulatory challenge. Traditional medical device regulation relies on predictable, repeatable outcomes. The ADVOCATE program is specifically designed to address this tension by incorporating real-time, AI-on-AI monitoring, which may eventually serve as a blueprint for how the FDA approves "learning" medical devices that evolve over time.
Implications: The Road Ahead
The implications of the ADVOCATE program extend far beyond heart failure. If successful, this project could provide the foundational "agentic" architecture for managing a variety of chronic conditions, from diabetes to chronic obstructive pulmonary disease (COPD).
However, the program faces significant hurdles. Public trust in AI remains a concern, and the "black box" nature of complex algorithms requires extreme transparency. Furthermore, the reliance on EHR integration—a historically fragmented and difficult-to-navigate space—remains a potential point of failure.
Despite these challenges, the shift toward agentic AI is clear. By automating the mundane and the critical, and by creating a safety net for rural patients, ARPA-H is betting that the future of medicine lies in the synergy between human clinicians and the digital agents that support them.
As the program moves into its active development phase, the medical community will be watching closely. If Tempus, Updoc, and their partners can satisfy the rigorous demands of the FDA while proving the efficacy of their agents in rural settings, they will have successfully bridged one of the most persistent gaps in modern medicine. The "clinical team" of the future may well be a hybrid of doctors, nurses, and the sophisticated algorithms that never sleep.
