FDA official: generative AI medical tools may be evaluated like physicians, not devices
nlarusstone · x · 2026-10-22
Rick Abramson, radiologist and director of FDA's Digital Health Center of Excellence, said at the AdvaMed MedTech Conference that generative AI medical tools may need to be assessed more like a physician's ongoing competency reviews than traditional software.
- The agency has spent 20 years evaluating "software as a medical device" under a simple input-X-output-Y paradigm (e.g., a chest X-ray AI giving a binary pneumothorax signal), which is evidentially straightforward.
- Generative AI, he argued, is starting to "behave a little bit more like me as a physician than a heart valve or a defibrillator," requiring a physician-style evaluation framework instead.
- Industry observers note this could also open a path for deploying useful AI agents in GxP-regulated environments.
More from Safety
- moyix: by the time you import the malicious package, arbitrary code has already run — moyix · 2026-10-22
- repligate Challenges the Orthogonality Thesis: Smart Systems Can't Have Arbitrary Traits in the Real World — repligate · 2026-10-22
- Security researchers warn simultaneous PHP/GitLab vulns make next 8 months 'continuous hell' — joshua_saxe · 2026-10-22
- Can defenders use the same LLM to find and patch an undisclosed PHP exploit first? — ccerrato147 · 2026-10-22
- AISI backs Anthropic's Mythos: only model to finish a 32-step corporate network attack sim — dbreunig · 2026-10-22
- Security researchers outbid Wiz for phpbleed.com to protest vulnerability credit-grabbing — dyn___ · 2026-10-22