FDA floats a two-axis risk framework and competency-style testing for generative AI medical devices; comment window closed Oct 19
On August 18, 2026, the FDA's Digital Health Center of Excellence (within CDRH) issued a discussion paper on how it might regulate generative AI-enabled medical devices, and opened docket FDA-2026-N-7874 on Regulations.gov for feedback through October 19, 2026. FDA says the paper is for discussion only: it is not draft or final guidance and does not set expectations for future marketing submissions.
The paper sketches a two-axis risk framework: one axis is how independently the device acts (from non-directive information, to output that directs an action, to fully autonomous function), and the other is how severe the harm would be if someone relied on a wrong output. FDA says it is considering treating "action-directing" GenAI functions as higher risk than ones that only provide information. For premarket review, it describes a competency assessment modeled loosely on how physicians are trained and evaluated, combining non-clinical benchmarking with clinical confirmation. It also outlines risk-proportionate postmarket monitoring options and raises questions about foundation models and agentic AI systems.
Practical upshot: nothing is binding yet, but if you build or buy GenAI tools that recommend or trigger clinical actions, expect FDA's eventual approach to scale scrutiny with how directive the output is and how much harm a wrong answer could cause, and to expect ongoing performance monitoring after launch.