What Is Project Perception?

Project Perception, currently in preview, is a Security Copilot agent-based infrastructure (referred to by Microsoft as “agentic defense”) intended to automatically detect and respond to issues it discovers. Using a team of red (attacker), blue (investigator), and green (responder) agents, Project Perception is intended to expose weaknesses that organizations might miss—or not catch as rapidly—and respond to them. (The colors are not arbitrary; they reflect a set of terminology common in the cybersecurity field.)

Project Perception consists of the set of agents, purpose-built models (MAI-Cyber-1-Flash among them), a degree of organizational context, the collection of Defender XDR and Sentinel’s signals and sensor data, actuators that determine how to respond to detected issues, and an orchestration harness.

Microsoft has not detailed the precise depth that green team agents are able to delve into when resolving issues, so it is unclear whether this is as simple as generating an alert for a human to respond to or if the agents will attempt to repair the issues. The types of issues these agents will be able to resolve are also unknown.