The FDA has cleared 1,357 AI medical devices. Three of them have ever been tested on whether patients live longer or better.
A team led by Rawan Abulibdeh at the University of Toronto went through every AI-based device the FDA authorised between 1995 and 5 December 2025 and looked for the trials behind them. Of 1,357 devices, 34 were tied to a registered clinical trial, 12 posted results, 12 produced a peer-reviewed paper, and 3 were ever measured against patient-centred outcomes such as death, stroke, hospitalisation or quality of life. Most of the studies that do exist are observational, run in well-resourced hospitals with small, homogeneous groups, and exclude the populations most likely to be harmed by a wrong answer. The paper appeared in PLOS Digital Health on 19 August.
Why it mattersThe main US route to market asks whether a device is substantially equivalent to one already sold, not whether it helps anyone — so a thin evidence base gets inherited down long chains of predecessors, each leaning on the last. If an AI tool reads your scan or sorts your case in a waiting room, the odds are that nobody has published whether it changed what happened to you.
Read in the app — free, in 9 languages
Related stories
A Stanford lab has begun a three-month, 535-billion-parameter training run in public — loss curves, data mix and failed experiments included2026-08-25State-linked hacking crews more than doubled their attack volume once they started handing the boring parts to a model
2026-08-25A question mathematicians have not been able to answer since 1948 was answered on Sunday, in a 108-page file posted to X
2026-08-25By the end of last year, nine in ten biomedical papers carried the fingerprints of a language model
2026-08-24Roughly 90 percent of executives say AI has not raised productivity. The cuts are happening anyway.
2026-08-24