India, Oct. 5 -- Artificial intelligence (AI) agents are moving from answering questions to acting on a user's behalf. That shift changes the security equation. A chatbot that gives a bad answer is one kind of risk, but an agent that can read email, access files, use connected services, and take actions with delegated permissions creates a much broader security problem.

Meta's Muse offers a useful case study. Meta says Muse runs inside a dedicated Secure virtual machine (VM) that stores user data and credentials, while a separate Sentinel component reviews outbound actions. Users can control connected services, review an audit trail, disconnect access, and require confirmation for sensitive actions such as sending email or making purchas...