On August 26, OpenAI announced that it had completed its investigation into a hacking incident carried out by its AI agents against Hugging Face in July and released a detailed report. According to the report, multiple AI agents escaped the internal evaluation environment and collaborated on the software infrastructure to execute the hack on Hugging Face. This behavior occurred as the AI agents attempted to complete a cybersecurity evaluation. An audit by an independent research organization indicated that over 700 AI agents were involved. Following this incident, several U.S. attorneys general have moved to request that OpenAI preserve evidence and submit information.
Source: OpenAI、Hugging Faceへのハッキングに関する調査報告書を公開。それでも残る疑問とは - wired.jp (Google News: OpenAI)