Zenity 称运行时分析只能发现智能体异常行为,无法阻止
AI 导读
Zenity 发文称,运行时分析只能发现智能体的异常行为,无法阻止其执行;预防需要接入能在动作运行前将其阻断的控制。该帖还称,数据外泄发生后的检测不等于安全。
正文 · 原文
Runtime analytics can tell you an agent did something unusual. It can’t stop it. 🔎
Preventing an action takes a connected control that blocks it before it runs.
Detection after exfiltration is not security.
👉 https://zenity.io/academy/runtime-ai-analytics
#AIAgentSecurity