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

People Affected

Large-scale AI incidents filtered by population impact. This visualization shows incidents where more than a threshold number of people were either exposed to risk or reportedly harmed, filtered by the highest direct severity score across harm categories.

How to read this

Each bar represents a year, stacked by risk domain. Filter by minimum severity (highest direct severity score across categories), population threshold (>1,000 or >1 million people), and impact type (exposed to risk vs harmed). Toggle between count (absolute numbers) and proportion (relative share per year). Click any bar segment to view matching incidents.

Key Takeaways

1.235 incidents involved more than 1,000 people exposed to risk.
2.Malicious Actors & Misuse dominates large-scale incidents at 34% of the total (79 incidents).
3.The top 3 domains (Malicious Actors & Misuse, Discrimination & Toxicity, AI System Safety, Failures & Limitations) account for 75% of all large-scale incidents, indicating concentrated risk.
4.Large-scale incidents decreased by 33% from 2024 to 2025.

Discrimination & Toxicity

Privacy & Security

Misinformation

Malicious Actors & Misuse

Human-Computer Interaction

Socioeconomic & Environmental

AI System Safety, Failures & Limitations

Source data from the AI Incident Database and MIT AI Risk Repository. Incidents prior to 2015 not shown.