AI diagnostic errors increase hospital blame unless doctors intervene

Researchers published a two-study vignette experiment in npj Digital Public Health examining how members of the public respond to adverse events in which AI played a role in diagnosis, per the journal article and News-Medical coverage. The News-Medical summary reports that participants attributed more responsibility to hospitals and were more likely to consider filing a complaint or pursuing legal action when AI was involved in a missed diagnosis. The first study included 299 online participants. News-Medical reports that negative reactions persisted even when an endoscopist reviewed the AI output, but responsibility attribution was significantly lower when physicians were substantively and interactively involved in the decision-making process.

Industry-pattern observations: research on human-in-the-loop medical AI frequently finds that perceived human oversight quality matters more than the mere presence of a clinician. Studies using vignettes typically measure attribution, trust, and intentions to complain or sue; the paper follows that design and reports those outcome measures for the experiments described. The vignette methodology captures hypothetical attribution rather than real-world litigation outcomes, which is an important limitation noted in News-Medical's coverage.

For hospitals and health systems, public attribution of blame carries reputational and legal consequences distinct from model performance metrics. Observers of health-technology deployment note that governance, documented clinician engagement, and transparent workflows tend to shape public perception and medicolegal exposure, even when technical failure modes are the proximate cause of harm.

For practitioners: observers and risk managers should track three indicators in deployments where AI assists diagnosis:

• •how clinician review is described and documented in patient records
• •patient-facing communication about AI's role in care
• •early complaint or litigation patterns following adverse events. The publication in npj Digital Public Health provides a citable peer-reviewed basis for governance and liability discussions

The findings derive from vignette experiments and hypothetical scenarios; real-world behavior and legal outcomes may differ. Effect sizes and full methodology are available in the journal paper.