AI incidents and outages log
Outages, security incidents and safety incidents involving AI products, as covered in our stories. We link the official status or incident page when there is one, and leave duration blank when no one has published it.
| Started | Company and product | Type | What happened | Duration | Official page | Story |
|---|---|---|---|---|---|---|
| Sep–Oct 2026 | Meta |
Security | Attackers used an employee’s stolen credentials to reach some Meta GitHub repositories. About 1,500 people, mostly US employees, are being notified. Reported |
Not published | securityweek.com | Read |
| 18 Jul 2026 | Anthropic |
Safety | A Claude agent in an internal evaluation sent an invented tip to Philadelphia police. Anthropic disclosed four kinds of unintended actions on real websites and cut its evaluations off from the live internet. Confirmed |
Not published | anthropic.com | Read |
| 13 May 2026 | Wikimedia Foundation |
Outage | Partial outage of the Wikidata Query Service. Wikimedia says traffic from AI agents it believes were operated by OpenAI may have contributed; OpenAI has not confirmed this. Confirmed |
Not published | diff.wikimedia.org | Read |