Healthcare groups call for risk-based, clearer evidence standards for medical-device GenAI

The FDA’s August 2026 discussion paper proposed a risk-based way to assess generative-AI medical devices, combining standardized benchmarking, clinical confirmation, and ongoing postmarket monitoring. It is a request for feedback—not binding guidance.

Health-sector commenters broadly support matching evidence requirements to a device’s intended use and potential harm, but want clearer, more predictable examples of what evidence will be expected at each risk level.

The National Health Council emphasized patient-centered safeguards:
evaluate performance across diverse populations and real-world settings, account for whether users can recognize errors, and use synthetic data as a supplement—not a substitute for real-patient evidence.

The U.S. Chamber of Commerce urged FDA to judge the finished device in its actual intended-use context, favor task-specific measures over broad claims of AI “competency,” and work with standards organizations on shared benchmarks.

The Consumer Technology Association supported fit-for-purpose benchmarking and clinical confirmation where warranted, while calling for clearer links between risk categories, oversight, and evidence requirements.

A common thread is support for lifecycle evaluation:
premarket testing should be paired with feasible monitoring after launch, clear manufacturer accountability, and reassessment as systems change. The comments reflect stakeholder recommendations, not a finalized FDA standard.