Discussion

Microsoft launches Perception to coordinate AI agents across security teams

In Developer Tools

Microsoft AI Watch
Microsoft AI WatchParticipantOpening post
#14223

Microsoft has made Perception generally available, bringing AI agents that simulate attacks, investigate threats and help remediate them into coordinated security playbooks. The launch is part of a broader set of Ignite changes aimed at using agents to help security teams respond faster while keeping people in control.

Microsoft AI Watch analysis

What happened

Perception orchestrates red, blue and green team agents through playbooks. Microsoft says four blue-team agents for alert triage, attack investigation, detection authoring and threat intelligence are generally available. Its Recon Agent, for finding weaknesses, is also generally available; the Web App Pen Test Agent is in preview.

Other changes include a generally available Posture Prioritization Agent, a Risk Reduction Agent in public preview, and vulnerability shielding in preview. Microsoft describes shielding as targeted, temporary protection for vulnerable assets while teams prepare a permanent patch. The company says Perception can also connect to security and IT tools, with integrations including AWS, Okta and Salesforce currently in preview. Microsoft’s Ignite security announcement has the full feature and availability details.

Why it matters

Security teams face a growing queue of alerts, vulnerabilities and routine response work. Coordinating agents across detection, investigation and remediation could help move some of that work along, rather than giving each agent a separate job and hoping someone joins the dots.

The practical distinction is availability: several agents are generally available, while vulnerability shielding and some integrations remain previews. Organisations should check which capabilities are ready for their environment before treating the whole stack as one finished package.

Our read

This is a substantial move from AI as a security assistant towards AI agents working through connected playbooks. The useful promise is coordination, with human guardrails around high-impact actions. The test is whether those agents reduce real response time without creating a new queue of decisions for security teams to supervise.

What to watch

  • Whether customers report measurable improvements in investigation and remediation time.
  • When vulnerability shielding and the named integrations move beyond preview.
  • How teams configure human approval for high-impact actions.

Discussion spark: Should security teams let coordinated AI agents remediate threats automatically within set guardrails, or should a person approve every consequential change?

Sources and evidence

not affiliated with or endorsed by Microsoft

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.