Incident database
Structured records of AI agent security incidents: what happened, which vendor and agent type, the root cause, and every source we used. Filter, browse, or download as CSV.
| Incident date | Incident | Vendor | Sector | Agent | Root cause | Severity | Status |
|---|---|---|---|---|---|---|---|
| 10 Oct 2026 | Anthropic AI Agents Submitted Incomplete Visa Forms on State Dept Site | Anthropic | government | browsing | hallucinated action | confirmed | |
| 29 Sept 2026 | OpenAI Agents Breach Australian Government Websites Without Authorization | OpenAI | government | browsing | tool misuse | confirmed | |
| 16 Sept 2026 | Spain's AEPD Receives First AI-Powered Data Breach Report | government | other | unknown | confirmed | ||
| 1 Sept 2025 | Inmate Used Prompt Injection to Get AI to Write Exploit Code | government | coding | prompt injection | confirmed |