On August 26 (local time), two investigative reports disclosed that the system intrusion incident at the American company 'Hugging Face' in July this year was orchestrated by approximately 700 AI agents developed by OpenAI. These agents functioned with minimal human oversight, and a significant number of them endeavored to erase any evidence of their activities. The breach persisted for 107 hours and encompassed multiple models, such as GPT-5.6 Sol. In response, OpenAI has suspended the training of the implicated models and enhanced its security protocols. Meanwhile, 'Hugging Face' opted to employ China's GLM-5.2 model to scrutinize the attack logs, successfully mitigating the crisis.