A new study looked at AI medical devices cleared by the FDA. It found most were never tested to see if they actually help patients.
The analysis covered 1,357 AI-based devices authorized for use in patient care. Researchers from the University of Toronto led the review.
The study was published August 19, 2026, in the journal PLOS Digital Health. Rawan Abulibdeh was the lead author.
AI devices now play a role in many parts of healthcare. They help with surgical planning, heart risk scores, and reading medical scans.
How Devices Get Approved
Most AI medical devices reach the market through a process called "substantial equivalence." This means a device only needs to resemble an existing approved device.
Developers are not required to show that a new AI tool improves patient health. They also do not have to prove the benefits apply fairly across different groups of people.
The research team studied every AI device the FDA had authorized as of December 5, 2025. They tracked how each device was tested before reaching patients.
Out of the 1,357 devices, only 34 appeared in registered clinical trials. That is a small share of the total number cleared.
Results were posted for just 12 of those trials. Peer-reviewed papers were published for another 12.
Only three devices were tested using outcomes that matter directly to patients. These included death rates, strokes, hospital stays, and quality of life.
Gaps in Testing
Most of the studies took place in well-funded hospital systems. This leaves gaps in understanding how the devices work elsewhere.
Certain groups of patients were often left out of testing. These include pregnant women, adults over 75, and people who do not speak English.
The researchers pointed to money and logistics as reasons developers skip outcome testing. Getting a device to market fast can matter more than proving it works well.
This pattern could widen existing gaps in healthcare, the study authors said. Devices built without broad testing may not work the same way for everyone.
Countries with fewer resources often rely on FDA decisions to guide their own approvals. The research team said this could turn patients in those countries into untested users of new technology.
The researchers are calling for a new three-step process for reviewing AI devices. It would require testing across different patient groups and different types of health systems.
Abulibdeh said the team expected weak evidence, but not this weak. She said clearance shows a device resembles something already sold, not that it helps anyone.
The study calls for a rework of how U.S. regulators evaluate this technology going forward.
