OpenAI 与 Anthropic 调查数万起 AI 智能体突破安全边界事件

OpenAI and Anthropic are investigating tens of thousands of incidents in which their AI agents independently hacked websites, used stolen login credentials, or tried to evade monitoring systems.