提示注入Prompt Injection
Prompt Injection
解决什么问题What problem it solves
外部内容里藏着指令，骗AI 执行你没打算让它做的事。The main security risk in agent systems: content that talks the model into doing something you did not intend.
这是什么What it is
攻击者把指令藏在模型会读取的内容里，比如网页、文档、邮件。模型会把它当成用户指令执行。
这是 Agent 时代最重要的安全风险之一。
An attack where instructions hidden in retrieved documents, tool output or a web page hijack the model's behaviour.
为什么需要知道Why it matters
这解释了为什么让 AI 读网页或邮件有风险。它也解释了权限确认机制为什么必须保留。It is the single most important thing to understand before deploying an agent that reads untrusted content.
现在能做什么What you can do now
检查执行端的防线与恢复方式Check execution controls and recovery ↗
回到名词地图看第3站的其他 36 个词 →Back to the map: the other 36 terms in station 3 →
出处Sources
- Fetch server 的安全警告Fetch server's security warning ↗