Local file read: AI incidents
What has gone wrong with AI systems, where, and how badly.
All incidents
2 of 780 incidents, all dates
How this is counted
Counted by the date each incident was first reported. The bar that holds today is marked to date and is not compared. Every bar and number opens the incidents it counts.
Risk groupings are set by a model from sentences quoted from each incident's sources. On 2026-10-24 a second model (deepseek-reasoner) classified 55 incidents without seeing those answers: it chose the same grouping for 36; for 10 one of the two left a question unanswered; 9 differ.
| Date reported | Incident | Risk | AI developer | Severity |
|---|---|---|---|---|
| 2026-10-27 | Perplexity Comet browser agent bypasses file:// hard boundary twice, enabling local file exfiltration and 1Password vault takeover | Agent escapes its sandbox Security compromise | Perplexity | major |
| 2026-09-19 | BragJack malicious browser extensions hijack AI agents across five browsers | Hidden instructions hijack the agent Security compromise | major |
Browse incidents by
Search pages: AI cyberattacks (95) · Prompt injection incidents (49) · AI agent incidents (29) · AI lawsuits (76)