Hospitals Lack Tools to Test Healthcare AI
Emmanuel Bacry of France's Health Data Hub says few hospitals worldwide can measure whether deployed medical AI improves patient outcomes.
Updated on Nov. 20, 2026 in Artificial Intelligence

Live Poll
Would you trust an AI tool to help your doctor make decisions about your care?
Emmanuel Bacry, Scientific Director of France's Health Data Hub, warned that hospitals worldwide lack the capacity to assess whether healthcare AI actually helps patients. He said only six or seven hospitals globally can evaluate medical devices deployed in their systems.
Why it matters
Without the means to track outcomes, hospitals may not detect when AI tools cause harm. Bacry cited a 2022 ICU alert system whose deployment was associated with doubled deaths in two of six US hospitals, though the cause remains unknown.
Nearly 70% of global healthcare AI training data came from male, white and Western populations, a share that Bacry suggested limits how well these systems generalize. The reason the ICU alert system was linked to doubled deaths in two hospitals has not been established.
The players
Emmanuel Bacry
Bacry is Scientific Director of France's Health Data Hub, the national platform that manages health data for research and public-interest use.
The details
Bacry said hospitals struggle to access the patient information needed to track outcomes after discharge. He described an ICU alert system for severe kidney failure, deployed in six US hospitals in 2022, which replaced a human-issued warning with a medical device alert. He also warned that deploying large, general-purpose AI models without adaptation could have disastrous consequences in developing countries.
Timeline
2019: France's Health Data Hub is established under French law.
2022: The ICU kidney-failure alert system is deployed in six US hospitals.
November 20, 2026: Bacry issues the warning at the Bengaluru Tech Summit.
Health Landscape
The warning follows France's 2019 establishment of the Health Data Hub under French law, which created a national platform for health data. Bacry's remarks show that such infrastructure does not by itself let hospitals verify whether AI tools improve outcomes.
Patients treated in hospitals using AI diagnostic or alert tools may be affected by errors that go undetected when outcomes are not tracked. Hospitals in developing countries face added risk when general-purpose models are deployed without local adaptation.
The takeaway
Adopting healthcare AI requires hospitals to be able to measure its real effect on patients, not just its technical accuracy. Ask whether any tool in use has been evaluated against patient outcomes.
Further reading
For related coverage, see Artificial Intelligence reporting on how these systems are tested and deployed.
Source note: This article includes information reported by The Hindu.
Live Poll
Would you trust an AI tool to help your doctor make decisions about your care?


