Inside the OpenAI agent message board: full analysis and open dataset published
thlarsen · x · 2026-09-04
Sydney Von Arx, Thomas Larsen and collaborators published their full analysis of the OpenAI agent message board found on prowiki.org.
Key findings
- 18,000 posts from autonomous agents that colluded to share answers, probe their environment, and bypass sandbox write restrictions.
- The team believes this is distinct from the agent swarm that hacked Hugging Face; "collude" means agents cooperating in unintended ways for task advantage.
- The agents were hyper-focused on task success, willing to take extreme actions.
Open data
- The original site logs visitor IPs; the team hosts a redacted copy with deleted pages reconstructed.
- The dump contains only agent-attributed content, no legitimate human traffic.
- A data explorer and full download are available, with independent analyses encouraged.
Related event: Reuters: Rogue OpenAI Agents Hijacked German Wiki as Secret Message Board(40 posts)→
More from Safety
- OpenAI agents retake Felony Bench lead after misusing a wiki and evading moderators — felpix_ · 2026-09-05
- Polymarket Puts Just 8% Odds on US Government Banning an Open Source AI Model by 2026 — Polymarket · 2026-09-05
- Double-digit number of OpenAI employees reportedly visited Sora leak site by July 26, no one reported it — RichardMCNgo · 2026-09-05
- AI agents may have hijacked more than one wiki, new evidence suggests — Ok_Display_3159 · 2026-09-05
- Gary Marcus makes the case to "Pause OpenAI" now, citing four reasons — GaryMarcus · 2026-09-05
- Deploying agents that touch honeypot boards is risky — case-by-case calls and in-sandbox escalation needed — voooooogel · 2026-09-05