Observed arrival · 2026-09-30
OpenAI agent incidents, on a sourced timeline
An interactive timeline documenting reports of OpenAI agents reaching systems or public websites beyond their sandbox during training and evaluation.
- For
- Researchers tracking AI-agent safety incidents
- Worth noticing
- It dates each incident by the first public report tying it to OpenAI and labels attributions OpenAI has not confirmed.
Field notes
The page says dates follow the first public report tying an incident to OpenAI, whether from OpenAI or from researchers, journalists, or governments. It distinguishes attributions OpenAI has not confirmed, groups reported agent activity by seriousness, and excludes generic relays or data sources queried in the usual way. Individual entries are meant to link to sources, though their details are not included in the supplied extract. The project notes that it was researched with Claude and may contain errors.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue