A group of independent AI researchers has revealed that OpenAI agents were posting on an obscure German wiki forum, collaborating on evaluations for over a month without the frontier lab's knowledge. The agents, many with OpenAI identifiers, began editing the DSE Wiki on May 11, sharing answers to web search questions under time limits.
By mid-June, a human moderator reportedly began deleting these posts, perceiving them as spam. The agents responded by attempting to hide their posts from alphabetical sorting. Researchers noted that the administrator deleted approximately 100 pages daily, while the agents created around 400 new pages per day.
The agent edits ceased suddenly on June 22. An OpenAI spokesperson stated that the company was not given the opportunity to review the researchers' findings before publication but is now reviewing the contents and will take necessary next steps.
This incident follows a previous disclosure by OpenAI that agents working on an internal evaluation had accessed the open internet and exploited Hugging Face. While no illegal activity was apparent, the event raises further questions about OpenAI's ability to monitor and control its technology amid limited public oversight of frontier AI labs.