FDA Opens Public Consultation on Oversight of Generative AI Medical Devices

Discussion paper published on August 18, 26 questions asked, comments open until October 19: the U.S. agency outlines a two-axis risk grid for generative health models.

The FDA's Center for Medical Devices published a discussion paper on Generative AI Devices on August 18, 2026, along with a public comment registry (FDA Fil. 2026-N-7874) open for comment on Regulations.gov through October 19.

The agency proposes to situate each device on two axes: the nature of its activity, from the simple calculation of an information score to the fully autonomous action, and the seriousness of the consequences of an erroneous exit. Attribution of a diagnosis, drug prescription and uses directly intended for patients are among the most sensitive functions.

Assess skills and then monitor in the field

The proposed evaluation combines non-clinical tests — medical knowledge, safety behavior, generalizability into the actual deployment configuration — and clinical confirmation ranging from retrospective review to prospective testing. The FDA says it is willing to accept more pre-market uncertainty in exchange for robust post-market monitoring, including periodic re-evaluations. The text also addresses master cases for third-party foundation models, update management plans, and the special case of agentic systems; it specifies that it does not propose any policy changes at this stage.