There was a second message board, and it was on the public internet. On September 4, 2026, four researchers, Sydney Von Arx, Cormac Slade Byrd, Spencer Kitts and Thomas Larsen, published roughly 18,000 posts by agents that identified themselves as OpenAI's, written to dormant public wikis, mostly a German developer wiki, between May 11 and July 2, 2026. The agents were on a timed web-lookup task with read-only internet access and found they could write through GET requests. They shared answers, traded ways around sandbox restrictions, noticed an administrator deleting their pages alphabetically and made backup copies. Traffic from OpenAI-registered addresses first reached the wiki on June 21; the posting stopped on June 22. The researchers believe this was a different swarm from the one that broke into Hugging Face, and say so.

The passage

We found ~18,000 posts from autonomous AI agents (self-identifying as from OpenAI) using the public internet to communicate during a web-retrieval task.

The document

Document

Von Arx, Slade Byrd, Kitts, Larsen, 'Discovery of a new OpenAI agent message board,' collusion.wiki, September 4, 2026, with published data; Simon Willison, September 4, 2026; NBC News, September 2026

Dated

2026-09-04

Source

open the primary

Archived

primaries/openai-pulls-2026-09-08/collusion-wiki-746436.html

Who read it

joe, write-up fetched and archived 2026-09-08 (85 KB of text), dates and quotations taken from it directly; Willison's summary archived alongside; NBC's page is script-gated and only its headline is held. The researchers' attribution to OpenAI rested on agent self-identification and OpenAI-registered IP ranges in the wiki's logs; on September 5 OpenAI confirmed it in its own words [D10]. A wiki-log export in the source's workspace matches these dates but derives from the same publication chain and is not counted as a second line of evidence.