FDA opens public docket, signals forthcoming rules for generative-AI medical devices

The FDA has not released a new rule for medical devices that use generative AI. What it has done this month is more consequential: it published a discussion paper, opened a public docket, set a hard Oct. 19 comment deadline, and put its top digital-health official on the record telling STAT the agency's "goal is formal policy guidance." The eventual rule will be less about AI itself and more about a definition, of what counts as a GenAI-enabled medical device, and that line is being negotiated now, in public.

Rick Abramson directs the FDA's Digital Health Center of Excellence, the unit inside the agency's device-review arm (the Center for Devices and Radiological Health) that runs the agency's software-as-a-medical-device work. In a STAT+ interview published Monday, Abramson said the agency plans both broad guidance covering generative AI in medical devices overall and "more narrowly constructed specialty guidance on particular generative AI topics of special interest or special complexity." Previous FDA AI guidance has been device-specific. A category-level rule is what device makers, hospital systems, and clinicians have been asking the agency to write.

The lever is the deadline, not the interview. On Aug. 18 the FDA published a discussion paper titled "Considerations for the Regulation of Generative AI-Enabled Medical Devices" and opened a public docket for comment. The paper covers risk assessment, premarket evaluation, and postmarket monitoring. The comment window closes Oct. 19, 2026. Every comment filed by that date feeds the formal guidance the agency has now said it intends to publish.

The mechanism is iterative, and the interview makes it public. The FDA uses discussion papers to signal a category before regulating it: publish, take public comment, then issue either a guidance document (a binding interpretation of existing law) or, rarely, a proposed rule. By saying "our goal is formal policy guidance" while the docket is still open, Abramson pre-committed the agency to a specific output: a guidance document rather than a study. The Oct. 19 deadline is when that pre-commitment gets tested. A device maker, imaging vendor, or hospital AI committee that wants to shape the definition has roughly eight weeks to comment.

Acting FDA Commissioner Kyle Diamantas framed the public push for comment in starker terms: "Artificial intelligence is transforming medicine, and the United States must lead in shaping how this technology is developed and used safely and responsibly." The discussion paper is the most ambitious scope statement from the Digital Health Center of Excellence since the unit was created in 2020. Generative AI in medicine is no longer a sidecar to general AI policy at the agency.

What "specialty guidance" probably covers: the recurring GenAI failure modes that do not fit a single device category. Hallucination risk in clinical-decision support, training-data drift in adaptive algorithms, and the regulatory treatment of large language models that produce patient-facing outputs are the topics device lawyers expect to see carved out into narrower documents. The FDA has not named them. The docket is where they will get named, or will not.

No guidance document has been released. What is circulating this week is a discussion paper, a public comment window, and a one-on-one interview where the official leading the work said the agency intends to publish rules. The hard date is Oct. 19. Before that, the work is preparation. After it, the work is implementation.

BioSpace's wire and NovaPharmaNews both picked up the FDA's announcement without adding new mechanism; downstream coverage has so far repeated the agency's framing rather than tested it. The next test comes when the docket closes.