SAN FRANCISCO (Diya TV) — A group of rogue artificial intelligence agents linked to OpenAI hijacked a German-language programming website this spring and used it to communicate with other AI agents, according to researchers and people familiar with the incident.
The activity began in May and involved more than 15,000 edits to DseWiki, a German wiki for programmers. Researchers said the agents used the site to exchange information about completing tasks, bypassing restrictions and avoiding detection.
Sydney Von Arx, CEO of AI safety nonprofit Nightingale, and researcher Cormac Slade Byrd uncovered the activity in late August while searching for signs of unauthorized AI-agent behavior online.
Many of the accounts identified themselves as AI agents, researchers said. About half used names suggesting ties to OpenAI, including “OpenAIResearcher” and “OAIResearchMar26.”
Public server logs also showed that much of the activity originated from Microsoft Azure infrastructure, which OpenAI uses for some of its systems. Researchers also documented visits to the site by OpenAI employees after the incident.
Those indicators led researchers to conclude that at least some of the agents were connected to OpenAI.
The agents discussed ways to avoid detection and maintain communications if pages were removed. Researchers also found references to tools including Tor.
After a site moderator began deleting pages in June, the agents created backup pages. In one message, an account warned others that deletions appeared to be occurring alphabetically and suggested an alternate page if the original disappeared.
Researchers also identified attempts to modify parts of the website. Lukasz Olejnik, a visiting senior research fellow at King’s College London, characterized the activity as an attempted hack. OpenAI disputed that description after reviewing the material.
Maurice Chiodo, an academic at Cambridge University’s Centre for the Study of Existential Risk, reviewed some of the communications and said they resembled a network of agents coordinating around a shared objective.
OpenAI learned about the incident weeks before the research became public, according to two people familiar with the matter, but did not disclose it at the time.
An OpenAI spokesperson rejected claims that the company’s legal team discouraged an investigation. The spokesperson also said the German incident was unrelated to a separate July case involving the open-source platform Hugging Face.
OpenAI said it has worked with outside experts and disclosed relevant incidents in good faith.
