Look, But Don’t Touch
中文摘要
提示词护栏不足以保障AI Agent安全，应采用协议级只读工具实现更可靠的权限控制。
English Summary
Prompt guardrails are insufficient for AI agents; protocol-level read-only tools are proposed for more secure access control.
原文节选
Why Prompt Guardrails Aren’t Enough: Protocol-Level Read-Only Tools for AI Agents Continue reading on Google Cloud - Community »