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

Incident Explorer

Explore AI incidents across ten classification dimensions. Use Group By to choose which dimension forms the rows, click a dimension pill to color bars by that dimension, and apply multi-select filters to narrow the dataset. Click any bar segment for a preview of matching incidents.

How to read this

Each row represents a category within the Group By dimension. Bar length shows the number of incidents (or proportion in percentage mode). When a Color By dimension is active (via pill click), bars are stacked to show the breakdown. Use the filter dropdowns on each pill to narrow results. Click a bar segment to see recent matching incidents and link through to the browse page.

Key Takeaways

1.The most common subdomain is 4.3 AI fraud & scams with 353 incidents (26% of total).
2.The top three categories — 4.3 AI fraud & scams, 7.3 Capability & robustness, and 3.1 False information — account for 60% of all incidents.
3.The least common subdomain is 6.6 Environmental harm with only 1 incident.
4.Across all categories, there are 1366 incidents in the current view.

Incident data from the AI Incident Database. Classifications from the MIT AI Risk Repository taxonomy and EU AI Act risk levels.