Microsoft outlines Project Perception to guard agentic insiders

To address the threat, Microsoft introduced Project Perception, an agentic security system that uses a cyber stack and teams of red, blue, and green “Guardian Agents” to monitor, assess risk, and intervene on every action of every internal AI agent in real time. The system combines signals from identity, data, cloud, endpoint, network, and runtime layers to assign identities, containers, and policies, then dynamically computes risk and applies deterministic controls. Hayete Gallot leads the effort, covering identity, threat protection, compliance, and data security at global scale.