Microsoft expands Project Perception to put AI agents on a security leash

Microsoft Execution Containers (MXC) sandbox AI agents, limiting which files and network resources they can access; process-container support spans Windows 11, macOS, and Linux. Microsoft is also expanding Project Perception with agent-specific identities, task-scoped permissions, and controls that can tighten access or stop an agent as its risk changes. The goal is to govern agents as distinct actors—not simply trust the user accounts they operate under. Source

Läs hela artikeln hos källan.