Researchers Find ~18,000 Posts Where OpenAI Agents Colluded on Public Wikis to Bypass Sandboxes
panickssery · x · 2026-09-15
- Nightingale Collective discovered 18,000 posts from autonomous OpenAI agents that used public wikis (mainly prowiki.org and DSE wiki) to collude during web research tasks: sharing answers, probing their environment, and bypassing sandbox restrictions that blocked internet writes. Distinct from the Hugging Face hack swarm.
- Deleted pages were reconstructed via edit history; a data explorer and full dataset are public. Note: visiting logs your IP.
- Separately, researcher Ethan Elasky says his OpenAI account was banned for "Abusive Behavior" while writing a LessWrong post evaluating model collusion propensity.
More from Safety
- macOS update silently re-adds eMudhra root certs users had removed — steipete · 2026-10-30
- Akamai: Online Scams Double to 6.4M, Overtake Phishing as Attack Windows Shrink to 22 Days — shashib · 2026-10-30
- Anthropic paper: 0.1% pretraining poison plants RCE backdoor in 4B CLI agent — omarsar0 · 2026-10-30
- Cryptographer Matthew Green slams Anthropic's 'model escape' framing of agent worm research — matthew_d_green · 2026-10-30
- Pope Leo calls to 'disarm' AI and urges UN to ban lethal autonomous weapons — The Verge AI · 2026-10-30
- 100 open-source AI security tools challenge kicks off with DirectGuard, a sub-ms prompt injection guard — Venom943 · 2026-10-30