Home/Incidents/Harm Severity by Category

These visualizations are experimental and still being refined. Designs and data presentation may change.

Harm Severity by Category

How are severity scores distributed across different harm categories, and how have they changed over time? Each incident is scored on a 1-5 scale across 10 harm categories using the harm severity rating system. Source data from the AI Incident Database and the MIT AI Risk Repository.

How to read this

Use the Category dropdown to select a harm type. The Count view shows raw incident counts stacked by severity level (1-5) per year. The Proportion view normalizes each year to 100% to compare severity distributions independent of volume. The Average view shows a line chart of mean severity per year, revealing upward or downward trends. Click any bar segment to browse matching incidents.

Key Takeaways

1.1,366 incidents have harm severity data across 10 categories, spanning 2015 to 2026.
2.Toxic or Malicious Content has the highest average severity score at 1.65/5, suggesting it is the most consistently harmful category.
3.Toxic or Malicious Content has the most high-severity incidents (58 rated 4-5), making it the top category for extreme harm events.
4.Use the Average view to spot upward severity trends, and the Proportion view to compare severity distributions independent of volume.

Severity Distribution

1 Negligible

2 Minor

3 Substantial

4 Severe

5 Catastrophic

Severity scores from the AI Incident Database. Classification based on the harm severity rating system. Incidents prior to 2015 not shown.