FDA Issues Discussion Paper on Generative AI Medical Devices

The FDA's discussion paper addresses generative AI systems used in diagnostics, imaging, and treatment planning. It highlights challenges like model drift and data provenance that current 510(k) pathways do not fully cover. Unlike prior AI/ML guidance from 2021, this paper explicitly calls for lifecycle oversight and human-AI team performance benchmarks. The document draws on internal FDA reviews of over 500 AI-enabled devices cleared since 2020.

Context from related JAMA studies shows standalone AI sometimes outperforming physician-AI teams on narrow tasks, raising questions about how regulators should evaluate hybrid systems. Observational data from Epic's user group meeting indicate hospitals are already deploying generative tools for documentation with limited prospective validation. These trends suggest the FDA paper responds to real-world deployment outpacing evidence standards.

Next steps include a public comment period through late 2026 followed by draft guidance. The agency has signaled potential requirements for continuous performance monitoring when models exceed predefined drift thresholds. Stakeholders await details on whether generative AI will trigger new premarket pathways or remain under existing software-as-medical-device rules.