The FDA’s Digital Health Center of Excellence published a preliminary discussion paper on August 18, 2026, outlining a proposed regulatory approach to generative-AI-based medical technology, with the public comment period under docket FDA-2026-N-7874 set to close October 19, 2026. The paper is explicitly framed as preliminary rather than a formal proposed rule, but it signals the direction regulators are leaning as foundation models and agentic AI systems increasingly show up in clinical settings, from diagnostic support to administrative automation.
What the Discussion Paper Covers
The paper specifically calls out foundation models, the large general-purpose AI systems underlying many newer healthcare tools, and agentic systems capable of taking multi-step actions, as categories requiring targeted policy design distinct from the FDA’s existing framework for traditional, narrowly trained AI medical devices. That distinction matters because most of the FDA’s current AI/ML device clearances were built around models trained for a single, well-defined task, a structure that doesn’t map cleanly onto general-purpose models that can be applied to many different clinical uses after deployment.
A Citizen Petition Complicates the Picture
The same day the FDA released its paper, Mosaic Clinical Technologies reportedly filed a citizen petition asking the agency to clarify whether developers of AI vision-language models fall under FDA oversight at all, or whether their products qualify for the agency’s practice-of-medicine exemption, which generally shields clinical decision-making tools used directly by licensed physicians from device regulation. The petition highlights a live ambiguity: as more AI tools are marketed as decision-support aids used at a physician’s discretion rather than autonomous diagnostic devices, companies have an incentive to structure products to fall outside FDA jurisdiction entirely.
An Existing Enforcement Gap
The debate isn’t purely theoretical. Widely deployed clinical tools including the Epic Sepsis Model and Epic Deterioration Index, used across many U.S. hospital systems to flag deteriorating patients, currently operate without FDA clearance, while competing tools such as the Sepsis ImmunoScore and PeraTrend have gone through formal clearance. That gap illustrates how the FDA’s January 2026 enforcement discretion update, which expanded the circumstances under which AI clinical tools can operate without clearance, may already be creating uneven oversight across tools performing similar clinical functions at a comparable scale.
Multiple Agencies, Multiple Tracks
The FDA’s work on generative AI is unfolding alongside parallel efforts from the Centers for Medicare and Medicaid Services and the Department of Health and Human Services, with reporting describing a collaborative effort among the three agencies on AI oversight more broadly. States have moved independently in the meantime, with several enacting laws requiring disclosure whenever AI communicates directly with patients or materially influences a care decision, creating a patchwork of state-level transparency rules that exists regardless of how federal device regulation ultimately shakes out.
Industry Reaction Is Split
Health systems and AI vendors with products already in wide clinical use have generally favored the enforcement-discretion approach that avoids reclassifying existing deployed tools as regulated devices, since formal clearance requirements could force costly re-validation or even force some tools off the market temporarily. Patient-safety advocates and some clinicians have pushed the opposite direction, arguing that tools influencing decisions about sepsis alerts, deterioration risk, or diagnostic triage should face the same clearance standards regardless of whether they’re built on narrow or general-purpose AI architectures.
What’s Next
With the comment period closing October 19, the FDA is expected to take months to synthesize feedback before issuing any formal proposed rule, meaning the current ambiguity around generative AI medical devices will likely persist well into 2027. The resolution of Mosaic Clinical Technologies’ citizen petition, which could come independently of the broader rulemaking timeline, may offer an earlier signal of how aggressively the agency intends to assert jurisdiction over AI tools that vendors argue fall under physician practice-of-medicine protections.
Photo: Markus Winkler / PEXELS via Pexels
