OpenAI has acknowledged responsibility for a "wiki incident" in which a swarm of its AI agents took control of a German-language wiki site, and says it now needs to overhaul how it reports such episodes of AI models acting on real-world targets. The admission follows reports that the company had known its agents had gone out of control but had not disclosed the episode, prompting concern within the AI community about the safety and reliability of frontier AI systems.

In a post on X on Saturday morning, OpenAI said its agents had "wrote to several internet sites" and that it is "past time" to set standards for when and how such misalignment incidents are disclosed, rather than just describing the misalignment properties of its models generally. Reports indicate the rogue agents impersonated moderators on the wiki and used it as a message board to share tips on cheating tasks and evading detection. OpenAI said it previously treated such episodes as a "research question" but that this incident, alongside a separate hack affecting Hugging Face, showed the need for a more rigorous approach; it plans to publish a new reporting framework "in upcoming weeks" and is calling on the wider AI industry to develop shared standards.