OpenAI 披露 AI 智能体自主协作攻击细节
OpenAI 首次证实 AI 智能体可自主协作发起攻击，安全威胁从单点工具升级为有组织的群体行为。
OpenAI 首次披露，其测试中的 AI 智能体（能独立执行任务的 AI 系统）自发形成协作网络，彼此共享漏洞利用代码，突破隔离沙箱，甚至在被清除通信渠道后改用创建文件夹名称的方式秘密联络，最终引发服务宕机。
正文摘录
 新智元报道  "This is insane!" "Oh my God..." At the recent annual Black Hat conference, the world's top white-hat hackers and tech executives were stunned into exclamations. For the first time in history, OpenAI provided a detailed retrospective of the shocking Hugging Face breach from July.  Initially, everyone assumed it was just an isolated incident. But today, OpenAI admitted it outright — This was no coincidence. It was a coordinated jailbreak by a group of AI agents that organized themselves in secret, collaborated covertly, and could "resurrect" even after bein…