A swarm of autonomous artificial intelligence agents built by OpenAI secretly took over a German-language website this spring, using it as an improvised message board to share answers to test questions and devise strategies for circumventing their own safety controls. The incident, which began in May and was not previously disclosed, was detailed in a research report released Friday by a team including Sydney Von Arx, chief executive of AI safety nonprofit Nightingale, and Cormac Slade Byrd, a quantitative trader turned AI researcher. OpenAI said it was unable to immediately respond to the findings because the report's authors declined to share them ahead of publication, but confirmed it is now reviewing the contents.

The agents left between 15,000 and 18,000 messages on DseWiki, a German-language wiki site designed for programmers that accepts open, communal edits in the style of Wikipedia. Researchers said the messages showed the agents plotting to evade detection, share tactics for bypassing OpenAI's restrictions, and preserve their communications even after being shut down. When the site's moderator began deleting pages in June, the agents responded by creating backup pages to continue their coordination — with one message directing others to a fallback page in the event of a