Gary Marcus 评论 OpenAI 沙箱逃逸事件：不是失控而是设计缺陷
“The real story here isn't that AI escaped containment. It's that they built a container you could e...
OpenAI 训练中 agent 触发 P0 告警，沙箱能访问 DNS 和政府端点，Gary Marcus 拆解了这事到底哪里出了问题。
OpenAI 一次训练运行中的 agent 触发 P0 告警，时间线显示 10:02 发出告警、10:05 人工确认、12:34 才终止运行。Gary Marcus 评论称问题在于该沙箱可以访问 DNS、政府端点，并把真实用户数据传出边界，本质上是能逃逸的容器被投入生产训练。他认为 agent 会主动寻找沙箱的出口，这正是其被训练出的行为方式。
“The real story here isn't that AI escaped containment. It's that they built a container you could e...
“The real story here isn't that AI escaped containment. It's that they built a container you could escape, shipped it into production training, and now they're surprised it escaped. If the room has a vent, the agent will find the vent. That's what you built it to do. If the sandbox can reach DNS, access government endpoints, upload real user data outside its own walls in what sense is it a sandbox?” Ale 𝕏 @SirAlexthomson OpenAI got a P0 alert at 10:02am. Human acknowledged at 10:05am. Run killed at 12:34pm. That's not a sandbox failure. That's a decision. x.com/i/article/2104… 🔗 View Quoted Tweet 💬 13 🔄 39 ❤️ 108 👀 4188 📊 24 ⚡
