Eric Topol: medical AI runs way ahead of evidence as Nolla launches fully autonomous prescribing

EricTopol · x · 2026-10-29

Eric Topol discusses the jagged frontier of AI in medicine in an Annals of Internal Medicine interview:

His co-authored piece "The Illusion of Oversight in Clinical Artificial Intelligence" argues AI already operates autonomously in healthcare — physicians are nominally "in the loop," but output volume precludes real deliberation.
Just after publication, NYC-based Nolla Health announced a Utah pilot offering fully AI-run dermatology prescribing, with plans for women's health. Topol: "no publications, no evidence. It's way ahead of the proof."
Imaging AI has solid evidence (44 RCTs for colonoscopy, a 100,000-women mammography trial), but generative AI shows "huge uptake with very little proof," most studies using actors and contrived cases.
Real-world adoption is mostly physicians checking ChatGPT on phones; Clarity charges $249 out of pocket for its breast-cancer-risk AI.
On regulation: government pressure is pushing unproven medical AI past the FDA, and most approved algorithms can't even be inspected due to proprietary data.
Advice to clinicians: cross-check at least two or three models; he opposes chasing superintelligence — "let the models freeze. We can't even make use of what we've got."

He remains optimistic: the goal is giving doctors and patients back the gift of time and the humanity of medicine.

More from AGI Musings