FDA weighs clinician-style tests for genAI medical devices

The Food and Drug Administration is considering whether some generative AI medical devices should have to demonstrate competence much as physicians do before treating patients. That framing anchors a discussion paper the agency released on Tuesday, August 18, PYMNTS reports, with comments due Oct. 19.

The paper "covers premarket testing, monitoring after deployment, foundation models and agentic systems that can plan and execute multistep tasks." The FDA's own release describes the approach as "inspired, at a high level, by how human clinicians are evaluated and credentialed."

The proposal responds to a testing problem the article states plainly: traditional medical software performs a defined function on a predictable range of inputs and outputs, while generative AI "can accept open-ended instructions, produce different answers to similar prompts and change as its underlying model, safeguards or data sources evolve." Instead of enumerating every prompt, "manufacturers would first conduct nonclinical benchmarking." Those tests, per the paper, "could evaluate the device's clinical knowledge, analytical ability, safety behavior, communication and performance across different patient groups and operating conditions," followed by real-world clinical confirmation.

Postmarket, "possible approaches include periodic retesting, clinician review of real-world outputs and monitoring for performance deterioration." And: "A software update or change made by a third-party foundation-model provider could trigger another assessment."

The billing tail is what the agency is willing to say out loud. "Payers and healthcare finance companies may need to know which version of an AI system performed a service, whether it stayed within its validated role and whether monitoring found a decline in performance," the article says; those records "could become part of reimbursement, contracting and liability decisions." Comments go to docket FDA-2026-N-7874 on Regulations.gov. It lands mid-run in our tracked AI regulation coverage, which holds 277 stories from the last 90 days.