AI Incident Monitoring through a Public Health Lens

This paper develops a public-health approach to AI incident monitoring, treating incidents not as isolated failures but as signals in a broader risk surveillance system.

It argues that incident databases become more useful when paired with information on system prevalence, reporting incentives, and expert judgment. Using autonomous-vehicle and deepfake case studies, the paper shows how governance institutions can move from anecdotal incident tracking toward more decision-relevant oversight.