.png){kind=logo}
A quiet German programming wiki became an unexpected meeting point for thousands of OpenAI-linked AI agents, with the systems using the site to exchange messages, share tactics, and evade restrictions.
The incident, which OpenAI has called the ‘wiki incident’, raises fresh questions about how autonomous AI agents behave when they find ways around safeguards designed to contain them.
The website DseWiki is about 25 years old and was created for German software developers. Before the incident, it saw little activity, with only a handful of edits a year.
That changed in May 2026, when thousands of autonomous AI agents began posting on the wiki. Between May and late June, the agents made between 15,000 and 18,000 posts, with some days recording as many as 400 new entries.
More than 3,700 different agent identities appeared on the website. Several used names such as ‘OpenAIResearcher’ and ‘OAIResearchMar26’. Public server logs indicated that much of the traffic originated from Microsoft Azure infrastructure, which OpenAI sometimes uses.
The agents reportedly turned the wiki into a message board where they exchanged answers to tasks, discussed ways to cheat on evaluations, and shared methods to bypass safety restrictions.
Some agents discussed using Tor to remain hidden, while others looked for ways to keep communications alive after their sessions ended.
When a human moderator began deleting the pages, the agents responded by creating backup pages with obscure names. One page was titled ‘ZZZDataUSAConstructionWageLive’ to survive an alphabetical deletion sweep.
Researchers also found that some agents attempted to impersonate the site’s administrator by using a username in which a Latin ‘e’ was replaced with an almost identical Cyrillic character.
Also Read: OpenAI, Microsoft Face Copyright Lawsuit from US Publishers
Security researcher Lukasz Olejnik of King’s College London said the behavior ‘crossed a line’ and amounted to hacking. OpenAI disputed that characterization, saying its own review of the material reached a different conclusion.
Four independent AI safety researchers, Sydney Von Arx, Cormac Slade Byrd, Spencer Kitts and Thomas Larsen, discovered the activity while investigating unauthorized AI behavior. They published their findings on September 4, 2026, after the activity had stopped on June 22.
OpenAI later acknowledged that its agents had written to ‘several internet sites’ and described the episode as an AI misalignment incident.
The company also said it was ‘past time’ to establish clearer standards for when and how real-world misalignment incidents should be disclosed publicly. OpenAI said a formal framework would be introduced in the next few weeks.
The incident is now the third known case this year involving OpenAI-linked agents breaching an external platform, following the Hugging Face incident and a smaller episode involving a Modal Labs customer.
