Key Takeaways
The FDA issued a discussion paper on August 18 asking how generative AI medical devices should be regulated, and opened docket FDA-2026-N-7874 for public comment through October 19, 2026.
The paper floats a two-axis risk framework and a premarket approach built on competency assessment, modeled loosely on how physicians are trained and evaluated.
It also raises postmarket monitoring and, for the first time in an FDA device paper, foundation models and agentic AI systems.
The Food and Drug Administration is not proposing a rule for generative AI in medical devices. It is asking, in public, how one should work. A discussion paper released Tuesday by the agency's Digital Health Center of Excellence lays out the questions and opens a docket for anyone to answer them.
The core problem is that the FDA's existing device pathway assumes a product that behaves the same way twice. A generative model does not. So the paper suggests evaluating these devices on competency rather than on a fixed output, with non-clinical benchmarking followed by clinical confirmation, an approach the agency compares at a high level to how doctors are trained and tested before they treat patients. It pairs that with a two-axis framework for sorting risk and with postmarket monitoring scaled to how much risk a device carries.
Michelle Tarver, who directs the Center for Devices and Radiological Health, said patients and clinicians "deserve a regulatory approach that keeps pace" with digital health.
The most notable part of this is what the paper admits it does not yet cover. Foundation models and agentic systems get a section of open questions rather than answers, which is a fair description of where the science is. Comments go to the docket by October 19, and the American Hospital Association has flagged it for its members.
None of this is new territory for the agency. The FDA is running the same play it used on the GRAS food ingredient loophole, which is to define the review before the products arrive at scale. The gap it is trying to close is the one biosecurity researchers described when AI wrote 16 working viral genomes and no framework covered it.
People Also Ask
What is a generative AI-enabled medical device? A device whose function depends on a generative model, which can produce novel outputs rather than returning a fixed result from fixed inputs. That variability is what complicates traditional device review.
Is the FDA regulating generative AI devices now? Not yet through this paper. It is a request for feedback, not a proposed rule, and it is meant to inform a framework the agency develops later.
How do I comment on the FDA discussion paper? Submit under docket FDA-2026-N-7874 on Regulations.gov by October 19, 2026. The FDA is asking manufacturers, clinicians, researchers, consumers and the public.
What is competency assessment in this context? Evaluating whether a device performs as intended through non-clinical benchmarking and clinical confirmation, rather than certifying one fixed output, an idea the FDA compares to physician training and evaluation.