The AI Agent Security Flaw That Doesn't Require Hacking
Imagine your AI agent reads a private file and emails it to someone.
No hacked credentials. No forbidden tools.
Every action was technically allowed.
But you never asked it to do that.
This is the security gap we're overlooking.
原文：https://www.a8q9.cn/en/article/the-ai-agent-security-flaw-that-doesn-t-require-hacking-6b9204.html