The FDA’s Digital Health Center of Excellence published a discussion paper seeking public comment on how to regulate generative AI-enabled medical devices, with the comment period under docket FDA-2026-N-7874 closing October 19. The paper proposes a risk-proportionate regulatory approach — tailoring the level of scrutiny to how much potential harm a given AI tool could cause patients — reflecting how dramatically the AI-enabled medical device landscape has grown since the FDA first began clearing such tools.
A regulatory framework racing to keep up
The number of FDA-cleared AI-enabled medical devices has grown from fewer than 50 in 2020 to more than 700 by 2026, an explosion that has outpaced the agency’s original regulatory frameworks, most of which were designed around more static, traditional software-as-a-medical-device products rather than the generative AI systems now entering clinical use. Generative AI tools — capable of producing novel text, images or recommendations rather than simply classifying or scoring existing data — pose distinct regulatory challenges because their outputs are less predictable and harder to validate through traditional testing methods than earlier generations of medical AI software.
Why risk-proportionate regulation matters
The FDA’s proposed approach would calibrate regulatory scrutiny to a device’s potential for patient harm, meaning a generative AI tool that drafts clinical documentation for physician review might face lighter requirements than one that directly recommends treatment decisions without human oversight. That tiered approach mirrors long-standing FDA practice for traditional medical devices but represents a meaningful adaptation challenge for generative AI tools, whose capabilities and failure modes don’t always map cleanly onto existing device risk classification categories.
What’s driving the surge in clearances
The explosive growth in AI-enabled device clearances reflects both genuine clinical utility — AI tools for radiology image analysis, clinical documentation, and diagnostic support have demonstrated measurable value in numerous studies — and the competitive pressure among health technology companies to bring AI products to market quickly as healthcare systems increasingly expect AI-assisted tools as standard practice. Radiology has been a particular hotspot, with FDA data showing the majority of cleared AI-enabled devices concentrated in that specialty, where image-based diagnostic tasks have proven especially well-suited to current AI model architectures.
The industry’s stake in the comment period
Health technology companies developing generative AI products have strong incentives to weigh in during the comment period, since the resulting framework will directly shape how quickly and cheaply they can bring new generative AI tools to market — a lighter-touch framework favors faster, cheaper market entry, while a stricter one could slow deployment but potentially build greater clinician and patient trust in AI-generated clinical content. Patient advocacy groups and medical professional organizations are expected to weigh in as well, often favoring more conservative frameworks given generative AI’s documented tendency to occasionally produce plausible-sounding but inaccurate outputs.
The clinical trust question
Physicians and health systems have voiced mixed views on generative AI tools in clinical settings — valuing time savings on documentation and administrative tasks while remaining cautious about tools that generate clinical recommendations or patient-facing content without robust human review built into the workflow. That tension between efficiency gains and accuracy risk is precisely what the FDA’s risk-proportionate framework is attempting to formally codify into regulatory practice.
What it means next
With the comment period now closed, the FDA will need to synthesize feedback from industry, clinicians and patient advocates into a finalized regulatory framework — a process that will shape how quickly the next wave of generative AI medical tools can reach clinical practice. Given how fast the AI-enabled device count has grown over just the past six years, the resulting framework will likely need built-in flexibility to accommodate AI capabilities that continue evolving faster than most regulatory processes are typically designed to track.
Photo: Mikhail Nilov / PEXELS via Pexels
