UpDoc has been selected as a lead performer in ARPA-H’s ADVOCATE program, receiving an award of up to $9.2 million to develop agentic AI for cardiovascular care. The company will work with Microsoft, OpenAI and NVIDIA on a system intended to manage aspects of heart-failure care under clinical safeguards and pursue a pathway toward FDA authorization.
The U.S. government is putting serious money behind the idea that AI agents could eventually take on parts of chronic disease management—not just assist doctors with documentation.
UpDoc, an enterprise clinical AI company, has been selected as a lead performer in the Advanced Research Projects Agency for Health’s (ARPA-H) Agentic AI-Enabled Cardiovascular Care Transformation, or ADVOCATE, program. The company says the award could total $9.2 million, with Microsoft, OpenAI and NVIDIA supporting the UpDoc-led team.
The goal is unusually ambitious: develop an autonomous clinical AI system capable of helping manage cardiovascular disease, with an eventual target of FDA authorization and real-world deployment.
ARPA-H describes ADVOCATE as an effort to create an AI system that could provide around-the-clock specialty cardiovascular care and act as a “clinician-extender.” The program is focused initially on patients with heart failure or those recovering from myocardial infarction.
That puts the initiative well beyond the current wave of healthcare copilots.
Most generative AI deployments in healthcare today assist with relatively bounded activities such as clinical documentation, information retrieval, coding or administrative workflows. An agent that can continuously monitor patient information, reason about changes and recommend—or eventually execute—care-management actions presents a substantially different safety and regulatory problem.
The economic case is considerable. The American Heart Association estimated average annual direct and indirect U.S. cardiovascular disease costs at $422.3 billion based on 2019–2020 data. A 2024 AHA analysis also projected annual healthcare costs attributable to cardiovascular risk factors could rise from $400 billion in 2020 to $1.344 trillion by 2050.
UpDoc says its proposed system will focus on areas such as medication, diet and exercise management for cardiovascular patients. Its architecture combines large language models with a proprietary Clinical Intelligence System intended to constrain agent behavior, maintain physician oversight and provide safeguards around clinical decisions.
That architecture is important because simply putting a general-purpose LLM in front of medical records would not satisfy the requirements of a production clinical system.
ARPA-H’s own ADVOCATE framework reflects that concern. The program has three technical areas: a patient-facing clinical AI agent, a supervisory agent designed to monitor the first system’s safety and effectiveness, and infrastructure for integrating AI agents into healthcare workflows.
The program also has a built-in validation and regulatory component. ARPA-H says performers that progress through milestone evaluations can move through independent verification and validation, benchmarking against cardiologists, FDA Investigational Device Exemption clinical trials and large-scale studies integrated with electronic health records.
That makes ADVOCATE notable within the broader AI healthcare platform market. Rather than developing a model first and figuring out regulation later, ARPA-H is attempting to build technical, clinical and regulatory infrastructure in parallel.
The distinction matters because the FDA’s AI-enabled medical-device landscape is already large, but predominantly consists of more bounded systems. The agency maintains a public list of AI-enabled medical devices authorized for U.S. marketing, and its current database includes devices across areas such as radiology, neurology and cardiovascular imaging.
FDA officials said in 2025 that more than 1,000 AI-enabled devices had been authorized through established premarket pathways, while emphasizing the need for lifecycle controls, transparency, bias management and postmarket performance monitoring.
Agentic clinical AI raises another level of complexity because the system can potentially perform multiple reasoning and action steps instead of producing a single diagnostic output.
That is where UpDoc’s partnership structure becomes significant. Microsoft brings cloud and enterprise infrastructure, NVIDIA supplies AI computing infrastructure, and OpenAI provides access to advanced large language models. But none of those components, individually, solves the central clinical problem: deciding when an AI system is allowed to act, when it must defer to a clinician and how its behavior can be audited afterward.
UpDoc already has an FDA-cleared insulin-management system, according to the company, designed to operate within physician-prescribed treatment plans and defined clinical boundaries. The company says the broader architecture being developed through ADVOCATE has not been cleared by the FDA.
The distinction is worth emphasizing. Selection for ADVOCATE funding is not FDA authorization, and participation in later stages depends on competitive milestone evaluations.
ARPA-H’s program itself acknowledges that safety is paramount. Its solicitation asks teams to address unmitigated risks associated with agentic AI in healthcare and emphasizes interoperability, orchestration and clinical validation.
If successful, the project could become a reference architecture for a new category of autonomous healthcare AI systems: models that do more than summarize medical information and instead participate continuously in patient care.
For the AI industry, that is the more interesting development. The race is no longer only to build larger LLMs. It is increasingly about surrounding those models with domain-specific intelligence, supervisory agents, data infrastructure, workflow integration and safety mechanisms that make autonomous systems usable in high-stakes environments.
Healthcare may be one of the hardest places to prove that model capability can translate into trustworthy autonomy. ADVOCATE is effectively a government-backed attempt to find out whether it can.
Market Landscape
Healthcare is becoming an important proving ground for agentic AI, clinical AI platforms and AI-enabled medical devices. Microsoft, Google, NVIDIA, OpenAI and a growing group of health-tech companies are developing technologies around clinical documentation, decision support, patient engagement and care coordination.
The harder problem is autonomy. A clinical agent must operate with incomplete information, changing patient conditions and potentially serious consequences from incorrect actions. That shifts the competitive focus from LLM benchmark performance toward safety cases, monitoring, auditability, human oversight, interoperability and regulatory evidence.
ARPA-H’s ADVOCATE program is particularly significant because it explicitly connects agent development with FDA validation, EHR integration and real-world clinical studies.
The opportunity is also large. CDC data show that heart disease and stroke caused more than 850,000 U.S. deaths in 2024 and generated hundreds of billions of dollars in direct and indirect costs.
Top Insights
- UpDoc will lead a federally backed effort to develop agentic AI for cardiovascular care, with potential funding of up to $9.2 million.
- Microsoft, OpenAI and NVIDIA are supporting the project, combining clinical AI, cloud infrastructure, advanced models and accelerated computing.
- ARPA-H intends ADVOCATE to pursue FDA authorization, clinical trials and health-system deployment rather than stopping at an experimental prototype.
- The architecture separates patient-facing AI from supervisory safety mechanisms, reflecting the additional controls required for autonomous clinical systems.
- UpDoc’s next-generation platform remains investigational; its ARPA-H architecture has not been cleared or authorized by the FDA.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
