If they could give their AI Agent access, then by definition it wasn't a secure system.