The Three Priorities HHS Is Acting On
HHS Deputy Chief AI Officer Arman Sharma framed the entire effort around one word: trust.
“Trust in this technology is the only thing that will lead to responsible but also effective adoption,” he said.
That framing is deliberate. It tells you exactly where the bottleneck is — not in building AI tools, but in getting clinicians, health systems, and patients to actually use them.
The three priorities reflect that reality directly.
1. Better Agency Coordination
Right now, clinical AI sits at the intersection of multiple federal agencies — HHS, FDA, ONC, ARPA-H, and others — with overlapping and sometimes conflicting oversight roles. That fragmentation slows everything down.
HHS is pushing for tighter coordination across these bodies, and the FDA’s Digital Health Center of Excellence is already signaling it will align with other federal and international regulators. For AI tool vendors, this means the regulatory landscape is about to become more coherent — but also more demanding.
2. Implementation Guidance and Support
This is the priority that most directly affects AI tool adoption at the ground level. Sharma was explicit: even with smart regulation and aligned reimbursement, there’s still an “implementation gap”.
Providers need governance frameworks. They need workflow integration support. They need confidence that a tool has been properly evaluated before they deploy it in a small clinic or a large academic medical center. Building the tool is only half the job.
3. Clearer Evaluation Standards
The third priority is about establishing consistent, credible standards for assessing whether AI tools actually work — and whether they’re safe. This is the piece that’s been missing from the market for years.
Without shared evaluation benchmarks, every health system has to do its own due diligence from scratch. That’s expensive, slow, and inconsistent. Standardized evaluation frameworks will lower that barrier and accelerate adoption across the board.
What the FDA Is About to Do
Rick Abramson, director of the FDA’s Digital Health Center of Excellence, made it clear that new regulatory proposals are coming — and coming soon.
He outlined four areas the FDA plans to address:
- Clarity on what the FDA regulates and what’s required of sponsors in both premarket and postmarket settings
- Risk-proportional oversight — meaning lower-risk tools won’t face the same scrutiny as high-stakes diagnostic or prescribing AI
- Lifecycle oversight for AI products after they enter the market, not just before
- Cross-agency and international policy coordination to reduce regulatory fragmentation
The FDA’s acknowledgment that AI evolves faster than traditional regulatory frameworks is significant. It’s a public admission that the current system wasn’t built for this — and that they’re actively rebuilding it.
“The world is looking to the FDA for leadership in how to approach advanced clinical AI tools,” Abramson said.
Expect formal proposals for stakeholder comment in the near term.
ADVOCATE: The First Signal of What Agentic Clinical AI Looks Like
One of the most striking announcements came from ARPA-H, which is developing ADVOCATE — the Agentic AI-Enabled Cardiovascular Care Transformation program. If it clears FDA approval, it would be the first FDA-approved clinical agentic AI system in the U.S.
The scope is ambitious. ARPA-H program manager and cardiologist Haider Warraich described a system capable of handling everything a clinician can do over the phone — scheduling appointments, dietary guidance, diagnosis, triage, medication changes, and new prescriptions.
That’s not a chatbot. That’s a clinical agent operating in high-stakes territory.
Warraich was candid about the challenges. He called implementation “incredibly challenging” and advocated for a supervisory layer to maintain human oversight post-deployment. He also invited health systems to co-develop the technology and test it through randomized clinical trials.
This is the clearest preview yet of where agentic AI in healthcare is heading — and the governance questions it will force the entire ecosystem to answer.
What This Means for AI Tool Adoption in Practice
These policy signals have real consequences for anyone operating in the clinical AI space.
For AI tool vendors, the message is clear: regulatory clarity is coming, but so is accountability. Risk-proportional oversight means low-risk tools may move faster, but anything touching diagnosis, triage, or prescribing will face serious scrutiny. Build your evaluation documentation now, not after the proposals drop.
For health systems and providers, the implementation gap is finally being acknowledged at the federal level. That means more structured guidance is coming — but you don’t have to wait for it. Start mapping your workflow integration requirements today so you’re ready to move when the frameworks arrive.
For buyers and evaluators of AI tools, evaluation standards are about to get more formalized. That’s good news. It means you’ll eventually have clearer benchmarks to compare tools against, rather than relying solely on vendor claims or internal pilots.
For the broader AI tools ecosystem, this is a maturation signal. The era of “move fast and figure out governance later” in clinical AI is ending. The tools that will win in this environment are the ones built with implementation support, transparent evaluation, and lifecycle accountability baked in from the start.
The Bigger Picture
HHS’s three priorities — coordination, implementation support, and evaluation standards — aren’t just bureaucratic checkboxes. They represent a structural shift in how the U.S. government plans to manage one of the most consequential technology transitions in healthcare history.
The 7,000+ public comments that shaped these priorities tell you something important: the healthcare community is engaged, opinionated, and ready to move — if the right frameworks exist.
The frameworks are coming. The question is whether the tools in the market will be ready to meet them.
Observe closely. The next 12 months in clinical AI will separate the tools built for the long game from the ones that weren’t.
Comments (0) No comments yet
Want to join this discussion? Login or Register.
No comments yet. Be the first to share your thoughts!