Read-Only AI Agents Can Still Leak Data
中文摘要
只读AI代理仍存在泄露风险。安全核心并非写入权限，而是缺乏出口控制，必须限制数据的对外传输以确保安全。
English Summary
Read-only AI agents can still leak data. True security requires egress control instead of just restricting write permissions to prevent unauthorized data transmission.
Original Excerpt
The missing security layer is not write permission. It is egress control. Continue reading on Medium »