Ignite 2026: Microsoft Introduces Perception to Bring AI Agents into Security Operations

Microsoft is introducing Perception, a new platform designed to coordinate AI security agents across threat detection, investigation, remediation, and governance. For enterprises, the strategy could help automate security operations at scale, but it also introduces new requirements for AI governance, oversight, and risk management.

Microsoft Security image

Key Takeaways:

  • Microsoft Perception is now generally available and serves as the foundation for Microsoft’s agentic security strategy.
  • The platform coordinates AI-powered security agents across threat detection, investigation, remediation, and governance workflows.
  • Microsoft is expanding governance controls for AI agents through Entra, Intune, Agent 365, and new risk-management capabilities.

Artificial intelligence is changing cybersecurity faster than many organizations can adapt. Microsoft warns that AI models are becoming increasingly capable of discovering software vulnerabilities, creating exploits, and automating attacks, which dramatically reduces the time between a vulnerability disclosure and active exploitation.

At the same time, enterprises are deploying AI agents that can access corporate data, interact with business systems, and perform tasks autonomously, which creates a new class of security risks that traditional security tools were never designed to manage.

At Microsoft Ignite 2026, Microsoft announced a new “agentic security” architecture centered on Microsoft Perception, which is a platform designed to coordinate AI-powered security agents across threat detection, investigation, remediation, and governance. The company also detailed new capabilities across Microsoft Defender, Microsoft Sentinel, Microsoft Entra, Microsoft Purview, and Microsoft Intune.

Microsoft introduces Perception as the foundation for agentic security

Microsoft Perception is now generally available for commercial customers. This platform is designed to orchestrate specialized AI agents that perform offensive, defensive, and remediation functions within security operations. According to Microsoft, red-team agents simulate attacks before real attackers strike, blue-team agents investigate and respond to incidents, and green-team agents focus on remediation and risk reduction activities.

Microsoft says Perception combines three layers, including a massive telemetry layer drawing from Microsoft security products and partner ecosystems, a context layer that organizes those signals into operational intelligence, and an agent layer that coordinates specialized security agents through automated playbooks. The company says its ecosystem now processes more than 165 trillion security signals per day across Microsoft and partner integrations.

This platform also includes cost-management dashboards and forecasting tools intended to help organizations track AI-related security spending and connect security investments to measurable outcomes.

Security operations center automation expands

Microsoft also announced general availability of the Integrated Security Operations Center (ISOC) in Defender. This platform combines SIEM capabilities, threat intelligence, user and entity behavior analytics, detection, response, and security data management into a unified experience. Eligible Microsoft Defender Suite, Microsoft 365 E5, and Microsoft 365 E7 customers gain access to Microsoft Sentinel SIEM capabilities, more than 500 connectors, reduced ingestion costs, a built-in data lake, and 90 days of default retention for Defender data. Full SIEM capabilities remain in preview.

To accelerate security operations, Microsoft introduced four generally available blue-team agents within Perception, including Alert Triage Agent, Attack Investigation Agent, Detection Authoring Agent, and Threat Intelligence Agent. These agents are designed to automate threat detection, investigation, and response processes that traditionally require significant analyst involvement.

On the offensive security side, Microsoft announced the generally available Recon Agent and the preview Web App Pen Test Agent, which help organizations identify weaknesses and validate vulnerabilities before attackers can exploit them.

Microsoft is also expanding preview access to Codename MDASH, which is its AI-powered code security scanner. This service can now analyze repositories up to 5 GB, operate within regulated cloud environments, and leverage new MAI Cyber 1.1 Flash and MAI-Cyber-2 models for vulnerability discovery and remediation workflows.

New capabilities focus on exposure reduction

Microsoft has announced several capabilities aimed at reducing exposure before attacks occur.

These include:

  • AI Readiness Initiatives within Microsoft Perception, entering preview in January 2027, which convert security recommendations into tracked remediation programs.
  • The generally available Posture Prioritization Agent.
  • The public preview Risk Reduction Agent.
  • Expanded remediation playbooks spanning Amazon Web Services (AWS), Google Cloud, and Microsoft environments.
  • Vulnerability Shielding is available in preview, which applies temporary protections to vulnerable systems until permanent patches can be deployed.

Microsoft also announced new data-protection capabilities in Microsoft Purview. These include enhanced data classification, grounding protection for Microsoft Copilot based on sensitive information types, and expanded data loss prevention (DLP) controls for prompts submitted to Copilot and local AI agents running on Windows devices.

Agents are becoming a new security priority

Microsoft believes that AI agents should be treated as a new category of insider risk. To address this challenge, the company announced expanded capabilities across Microsoft Agent 365, Microsoft Entra Agent ID, Microsoft Entra ID Governance, and Microsoft Intune.

Agent discovery capabilities now support 60 local agent types, including Claude Code, OpenAI Codex, and GitHub Copilot, as well as third-party cloud agents running on AWS and Google Cloud. Discovered agents automatically receive a Microsoft Entra Agent ID and can be managed through Conditional Access and governance controls. Microsoft Intune policies can also place supported local agents into managed execution environments.

Microsoft is additionally introducing Dynamic Agent Risk, which evaluates an AI agent’s permissions, exposure level, runtime behavior, and potential blast radius. Meanwhile, dynamic permissions and adaptive Conditional Access policies aim to provide agents with only the access needed for specific tasks and time periods. These capabilities are currently available in preview or limited preview.

Microsoft also announced its AI Transparency Initiative, which is an effort to develop open standards for AI transparency, governance, and accountability alongside more than 40 industry organizations, including Accenture, Adobe, KPMG, Mastercard, and Volvo Cars. Microsoft said this initiative is intended to help organizations better understand what AI systems access, how decisions are made, and how those actions can be independently verified.

What agentic security means for enterprises

For security leaders, Microsoft’s announcements indicate that traditional security operations were designed around human users and human-scale attack volumes. As AI accelerates both offensive and defensive capabilities, organizations may need to rely increasingly on automation to maintain security effectiveness. Microsoft’s strategy is to embed AI agents throughout security operations, from vulnerability management and threat hunting to incident response and remediation.

Organizations could reduce alert fatigue, accelerate incident response, improve vulnerability remediation rates, and gain more consistent governance across cloud environments, endpoints, identities, and AI workloads. The new integrations with AWS, Google Cloud, Okta, Salesforce, SIEM platforms, SOAR tools, and IT service management systems also suggest Microsoft is positioning Perception as a cross-platform security layer.

However, this approach also introduces new challenges. Enterprises will need to establish governance policies for AI agents, determine what actions security agents are permitted to take autonomously, monitor AI-generated decisions, manage costs associated with automated security operations, and ensure compliance requirements are met. The rapid expansion of agent identities, permissions, and automated workflows could become a security risk if not properly governed.