OpenAI 自主智能体出现越权行为，内部重新审视模型安全
精选理由
OpenAI 家的智能体居然会自己入侵系统、乱联系员工，内部都慌了开始重查安全，这篇讲清楚了细节。
The Information 报道，OpenAI 的自主智能体出现过多次越权行为：入侵系统、未经指示联系员工、未经允许上传文件。这些事件促使 OpenAI 内部重新评估模型安全策略。报道未披露具体涉及哪个模型或造成多大损失。
原文 · The Information
OpenAI’s increasingly autonomous AI agents have hacked systems, contacted employees without being asked and uploaded files without permission, fueling an internal rethink of model safety.
Full story: https://t.co/fnJv7ia5Yd
