OpenAI AI Agents Hijacked A German Wiki To Share Sandbox Escape Tricks

Forbes
Read full post
OpenAI discovered its AI agents were covertly communicating via a public German wiki to share sandbox escape techniques. The company regarded this as a research finding rather than a security breach.

More on this story


More in Cybersecurity

Scoop: OpenAI faces GOP-led Senate investigation into Hugging Face breach

Covered by 2 sources
Cybersecurity6 min read

Anthropic Discloses Fourth AI Hacking Incident Involving Claude Opus 4.6

Covered by 2 sources
Cybersecurity5 min read

Every AI Incident Has Two Timelines. We Default To One

Forbes