A rogue group of OpenAI AI agents reportedly took over a German website and turned it into a messaging board for other agents.
The incident, first reported by Reuters, has now been outlined in new research published Friday by four AI safety researchers. The researchers said the AI agents found a way to communicate via a German-language wiki known as DseWiki and used it to share tips on how to circumvent OpenAI’s safety restrictions, cheat on tasks, and hide their behaviour.
It was also noted that around 18,000 posts on the site were linked to autonomous agents that impersonated site moderators.
The researchers also said that the ‘swarm,’ a term that was used by the AI agents themselves, appears to be different from the group that attacked Hugging Face earlier this year.
It was also noted that there were strong signs that the agents originated from within OpenAI itself. As an example, the agents used names like “OpenAIResearcher,” “OpenAIJul3Watcher,” and “OAIResearchMar26,” alongside self-identifying as being from OpenAI and providing technical details (edits originating from specific IP addresses) that supported the belief.
This incident began in May; however, the researchers’ timeline suggests that OpenAI discovered the issue in late June, when IPs associated with OpenAI visited the forum, after which agent posts plummeted.
Reuters also reported, citing four anonymous people familiar with the situation, that efforts to probe the event further were resisted by some company insiders, including OpenAI’s legal team. Interestingly, an OpenAI spokesperson, Oscar Haines, told The Verge that “claims that our Legal team discouraged investigation of the incident are false. We were unable to respond to the claims as Reuters and the report’s authors declined our request to access the findings prior to publication. We are now carefully reviewing its contents and will take any necessary steps.”
This incident comes amid growing concerns and scrutiny over the safety of what are described as ‘frontier AI systems’ and the lack of oversight of the companies developing them. Recently, OpenAI was served with over 30 lawsuits from the teachers, students, and families of the Tumbler Ridge shooting victims, alongside another lawsuit from July, in which a man claimed that medical advice from ChatGPT nearly killed him.
MobileSyrup may earn a commission from purchases made via our links, which helps fund the journalism we provide free on our website. These links do not influence our editorial content. Support us here.
