Microsoft has officially moved its agentic security platform, Perception, out of public preview following its debut as a project in July and subsequent integration into Microsoft Defender in August. Announced during the Ignite 2026 conference, the platform is designed to orchestrate autonomous AI agents to perform complex security operations, including detection, investigation, and remediation. The system utilizes specific playbooks to coordinate agents categorized into red, blue, and green teams, each tasked with distinct operational roles.
The blue team agents, which are now generally available, focus on the core components of incident response. These include tools for alert triage, attack investigation, detection authoring, and threat intelligence. Microsoft reports that the Alert Triage Agent is capable of managing approximately 80% of alerts within a standard Defender queue, a figure derived from an analysis of roughly 8 million alerts across the company’s security product portfolio.
Red team functionality has also expanded with the general availability of the Recon Agent, which is designed to probe for system weaknesses. Additionally, a Web App Pen Test Agent has been introduced into the preview phase. These red team components are intended to simulate attacker behavior to help security teams identify and address vulnerabilities before they can be exploited by malicious actors.
Green team operations, which focus on remediation and posture management, have also seen updates. The Posture Prioritization agent is now generally available, while the Risk Reduction agent remains in public preview. This Risk Reduction agent is designed to provide vulnerability shielding, offering temporary protection for assets when a permanent patch is not yet available. These green team playbooks are built to function across multi-cloud environments, including Microsoft, AWS, and Google Cloud.
To ensure operational control, Microsoft has emphasized that the platform requires human oversight. Security teams are responsible for establishing guardrails and providing final authorization for high-impact actions. Automated triggers are configured to launch specific playbooks only when an incident meets pre-defined criteria, ensuring that the autonomous agents operate within the parameters set by the organization.
Beyond the core Perception platform, Microsoft is expanding its ecosystem through new integrations. The company is currently previewing capabilities that connect Perception to existing Security Information and Event Management (SIEM), Security Orchestration, Automation, and Response (SOAR), and IT service management tools through APIs and the Model Context Protocol (MCP).
Microsoft is also addressing the emerging security risks associated with the proliferation of AI agents within enterprise environments. Recognizing that these agents can act as insiders with extensive access to data and tools, the company has introduced detection capabilities within Agent 365 and Entra Suite. These tools are currently in preview and can identify approximately 60 types of local AI agents, such as GitHub Copilot, OpenAI Codex, and Claude Code.
To manage these risks, Microsoft is utilizing Intune policies to maintain control over supported agents on enterprise devices. This move is part of a broader effort to mitigate the potential for unauthorized or rogue AI agents to compromise corporate environments, a concern that has intensified as organizations increasingly adopt agentic workflows for development and security tasks.





