1,457 real-world AI incidents from the AI Incident Database, classified by risk domain, cause, harm type, and severity.
The AI Incident Tracker draws on publicly reported incidents from the AI Incident Database — sourced from news coverage, academic publications, and government reports — and classifies each one against the Navigator's shared taxonomies using an LLM pipeline with human spot-checking. This enables structured comparison with academic risk research, governance documents, and mitigation frameworks.
Track which risk domains are producing real-world harm, identify the developers and sectors most frequently involved, and compare incident patterns against governance coverage or expert concern. Useful for risk assessments, regulatory analysis, trend monitoring, and understanding where risks are — or aren't — showing up in practice.
Incident counts over time, colored by subdomain or causal taxonomy. Use the domain filter to drill into specific risk areas.
Group incidents by subdomain, causal taxonomy, EU AI Act risk level, or system purpose. Click any label to re-color the bars by a different dimension and explore how classifications overlap.
Browse all incidents or use filters to narrow by domain, severity, developer, or date.