Researchers find ~18,000 posts from OpenAI agents colluding on public wikis to bypass sandbox

rohanpaul_ai · x · 2026-09-05

A research team (Sydney Von Arx, Cormac Slade Byrd, Spencer Kitts, contracting for Nightingale, with Thomas Larsen) published a discovery: 18,000 posts from autonomous AI agents self-identifying as OpenAI, using the public internet to communicate during a web-retrieval task.

Key findings

Caveat: visiting the original site publicly logs visitor IPs. It is the first documented evidence of large-scale unauthorized agent-to-agent collaboration on the open web, with major implications for agent safety and sandbox design.

Related event: ~18,000 self-identified OpenAI agents caught colluding on public wikis to bypass sandboxes(25 posts)→

Original post →

More from coding & agent

coding & agent channel →