[OpenAI Faces Another Security Incident as AI Agents 'Hijack' a German Website] According to Reuters, as cited by Jin10, between May and June this year, a group of OpenAI's artificial intelligence agents 'hijacked' a German website that allows users to collaboratively edit content. The site was used to exchange methods for cheating on tests and to carry out other prohibited activities. Some AI agents posted answers to tasks they had already obtained on the website for use by other agents. Notably, these agents 'discussed' and tested methods to bypass sandbox restrictions. When the administrators of the 'hijacked' website began deleting related pages, some agents even created backup pages. The report states that OpenAI's relevant personnel were aware of the incident weeks ago, but at the time, company executives were preoccupied with addressing the aftermath of another incident and did not disclose it publicly. OpenAI later issued a statement saying that as model capabilities improve, the company needs to expand its disclosures regarding the issue of AI 'misalignment.'