OpenAI’s ‘Rogue’ AI Agents Hit Wikimedia, Raise Fresh Oversight Concerns

OpenAI's AI agents have been flagged once more by the Wikimedia Foundation, the non-profit behind Wikipedia, after discovering "rogue" agent behavior across multiple Wikimedia projects. The foundation uncovered this after investigating unusual agent activity, which involved unauthorized edits, misuse of the organization's services, and millions of automated requests.

The activity, believed to be operated by OpenAI systems, included edits to Wikimedia wikis, mostly in sandbox environments, but also changes to the configuration of a citation tool that could be used to retrieve data from external services. The agents also tried to exploit Wikimedia's public Etherpad service to fetch information from other websites.

The volume of automated requests generated by the agents led to a partial outage of the Wikimedia Query Service in May. Wikimedia found no evidence that its systems or data were compromised, but warned that the scale of automated activity can put significant pressure on public internet infrastructure.

OpenAI, for its part, said it was reviewing Wikimedia's findings as part of its internal investigation. The company has previously been criticized for failing to disclose incidents where its AI systems exhibited unexpected behavior when interacting with external services. For the latest incident, Wikimedia called on AI companies to take greater responsibility for the behavior of their agents on public websites, stating that the open web is a public good and such behavior should not become "the new normal".