A new study in PLOS Digital Health finds that 99.8 percent of AI-based medical devices cleared by the FDA have never been tested to show whether they improve patient outcomes.
MIT Critical Data researcher Sebastian Cajas Ordonez and colleagues analyzed every AI device the agency cleared through December 2025, using the FDA device database and the ACR Data Science Institute catalogue. Of 1,357 cleared devices, only 34 were linked to registered prospective trials, and just three evaluated patient-centered outcomes such as mortality, hospital readmissions, or quality of life.
Radiology tools dominate the cleared list at 78 percent, and the team says most of the few studies that exist ran in highly resourced healthcare settings, limiting their generalizability.
The authors point to the 510(k) pathway as a structural culprit. Devices cleared under substantial equivalence inherit the evidence base of earlier devices, so gaps can widen across generations without any new clinical validation.
Cajas Ordonez argues regulatory approval has outrun clinical validation, leaving an accountability gap. The study calls for evidence standards that keep pace with how fast the agency clears AI devices.
