What the research found
The FDA's Digital Health Center of Excellence issued a discussion paper acknowledging that generative-AI-enabled devices may introduce risks unlike those of traditional software or earlier AI devices, and opened docket FDA-2026-N-7874 for comment until 19 October 2026. It sketches a two-axis risk framework, a premarket approach built on 'competency assessment' — non-clinical benchmarking plus clinical confirmation, inspired by how physicians are trained and examined — and several options for risk-proportionate postmarket monitoring, including for foundation models and agentic systems. CDRH director Michelle Tarver framed it as a transparent process intended to serve as a potential model for regulators elsewhere.
Generative AI is already in clinical settings while the framework for judging it is still being drafted in public — which is both the clearest statement yet that clearance is not proof, and an open door for providers to say what evidence they'd actually need.