Most-implicated AI systems
Incidents involving each AI system, not attacks by a company. Click a row to see those incidents on the globe.
Fairness rules
- Wording is always incidents involving [model], never "attacks by [company]". In most cases a human misused the model.
- Self-disclosed marks incidents the AI provider reported itself. Labs that publish threat reports would otherwise look worst for being transparent, so read that count as a sign of openness.
- Incidents with unknown attribution go in an Unattributed row and are never assigned to a guess.
- Default view: confirmed + reported incidents, all-time. Claimed-only incidents are excluded unless you choose "All confidence".
| # | System | Incidents | Confirmed | Self-disclosed | Avg sev. |
|---|---|---|---|---|---|
| 1 | OpenAI | 9 | 9 | 4 | 2.1 |
| 2 | Anthropic | 5 | 5 | 4 | 2.4 |
| 3 | 1 | 1 | 1 | 2 | |
| 4 | Meta | 1 | 1 | 1 | 2 |