Microsoft Moves Perception Agentic Security Platform to General Availability
Share
Microsoft has moved its Perception agentic security platform out of public preview, making the AI-driven coordination system generally available for enterprise defenders. The platform, which integrates directly into Microsoft Defender, uses specialised teams of autonomous agents to automate security operations, incident response, and threat hunting.
Unveiled earlier this year as Project Perception, the system organises AI agents into red, blue, and green teams. Red agents simulate attacker behaviour to identify weaknesses; blue agents manage detection and triage; and green agents focus on remediating identified risks.
Blue Team Agents Automate Alert Triage
Four blue team agents have reached general availability, covering alert triage, attack investigation, detection authoring, and threat intelligence. According to data released by Microsoft, the Alert Triage Agent is capable of handling approximately 80% of alerts in a typical security queue. This estimate is based on an analysis of roughly 8 million alerts across the company’s security product suite.
The Recon Agent, a component of the red team designed to probe environments for vulnerabilities, is also now generally available. A more advanced Web App Pen Test Agent has entered public preview to assist in automated application security testing.
Vulnerability Shielding and Multi-Cloud Support
The platform’s green team agents focus on improving security posture across diverse environments. The Posture Prioritisation agent is now generally available for Microsoft Defender for Cloud, AWS, and Google Cloud. Meanwhile, a Risk Reduction Agent has entered preview, offering the ability to implement “vulnerability shielding.”
This shielding feature provides temporary protection for vulnerable assets when a permanent patch is not immediately available, buying time for security teams to coordinate a formal rollout. Operations are governed by “playbooks” that allow human teams to set guardrails and approve high-impact actions.
Managing the Risk of Autonomous Agents
Microsoft also addressed the potential security risks posed by autonomous AI agents themselves. Because these agents require broad access to data and tools to function, they could effectively become a new form of insider threat if compromised or misconfigured.
To mitigate this, new features in Microsoft Entra and Agent 365 are being previewed to detect and manage over 60 types of local AI agents, including tools like OpenAI Codex and GitHub Copilot. These controls allow administrators to apply dynamic permissions and risk scoring to autonomous entities within the network.



