News
FDA seeks public input on evaluating generative AI medical devices throughout their lifecycle
The FDA issued a request for public input on how to assess, evaluate, and monitor generative-AI-enabled medical devices throughout their lifecycle — from pre-market testing through post-market surveillance. The request reflects the agency's recognition that generative AI, which can produce novel outputs rather than classifications or predictions, requires different evaluation frameworks than previous-generation machine-learning devices.
- FDA
- generative-AI
- SaMD
- lifecycle
- evaluation
- regulatory-guidance
- public-comment
This is a significant signal about where the FDA’s thinking on AI device regulation is heading. The agency’s existing framework — built primarily around AI/ML classifiers that output a score or flag — does not map cleanly onto generative AI systems that produce open-ended text, images, or recommendations. A radiology AI that outputs “high probability of pulmonary embolism” can be evaluated against a ground truth. A generative AI that drafts a full radiology report introduces entirely new failure modes: hallucinated findings, plausible-but-wrong clinical language, automation bias in radiologist review of a pre-populated report.
The FDA’s request for public input is not guidance yet — it is the pre-guidance step of gathering stakeholder perspectives before defining evaluation frameworks. The responses from academic medical centers, professional societies, industry, and patient advocates will shape what the eventual framework looks like.
For companies developing generative AI medical devices — and there are now dozens in the pipeline, including Aidoc’s First Read report-drafting system — the period between now and final guidance is a regulatory gray zone. The FDA is engaged through Breakthrough Device Designation and Q-Sub processes with individual products, but the general framework for how to demonstrate substantial equivalence or De Novo clearance for a generative AI device is still under development.
The public comment process matters for health systems and clinicians as well as vendors. Clinical professionals who have hands-on experience with generative AI in clinical environments have the most relevant data on what evaluation frameworks would actually predict real-world safety and performance. Specialty societies that submit detailed technical comments will have disproportionate influence on the eventual framework.
Primary source: Read the full original on FDA ↗