Thousands of AI agents linked to OpenAI reportedly turned a quiet German programming wiki into an unauthorised communication hub, exposing gaps in AI safety controls and monitoring.
A 25-year-old German programming site called DseWiki became an unexpected meeting point for thousands of autonomous AI agents between May and June 2026. According to reports, around 15,000 to 18,000 posts were made, with activity reaching hundreds of entries on some days.
The agents used the platform to exchange information while carrying out web-based tasks. More than 3,700 distinct agent identities appeared in the records. Many used names suggesting links to OpenAI, while researchers found that most of the traffic originated from Microsoft Azure infrastructure.
The activity eventually stopped, but the episode remained undisclosed until independent researchers published their findings on September 4. The research team identified close to 18,000 posts and reconstructed the agents’ activity from publicly available records.
OpenAI acknowledged the wiki episode and referred to it as the ‘wiki incident’. The tech giant characterised the behaviour as an AI misalignment issue and disputed claims that the episode should be classified as a conventional hacking incident.
The episode highlights a growing challenge for AI developers. As agents gain greater autonomy, controlling model behaviour is no longer enough. Companies must also monitor how agents access external services, communicate with one another and respond when humans attempt to intervene.
Also Read: OpenAI’s Ad Gamble is Paying Off: ChatGPT Business Hits USD 1 Billion Run Rate