Microsoft has made Perception, its agentic security platform, generally available at Ignite 2026. The platform orchestrates red, blue and green team agents through playbooks, letting AI defenders triage alerts, hunt threats and fix flaws at machine speed across the network.
What Perception does
Perception brings the top of Microsoft’s security stack together. Red team agents simulate attacks before a breach, blue team agents detect and investigate them, and green team agents fix what is broken, all working in a closed loop through shared playbooks.
Four blue team agents are generally available now: Alert Triage, Attack Investigation, Detection Authoring and Threat Intelligence. The Recon Agent and the Posture Prioritization Agent are also generally available, while the Risk Reduction Agent and the Web App Pen Test Agent remain in preview.
Microsoft says the agents free analysts for the decisions that matter most. Teams set the guardrails and sign off on high-impact actions, while automated triggers launch playbooks when an incident meets the criteria they set in advance for their environment.
Why Microsoft built it
The launch answers a shift Microsoft says is already here: frontier models can now find and exploit software flaws, and open-weight models with limited guardrails are putting that capability within reach of almost anyone. Attacks that once needed specialist teams now take about a day.
The numbers behind the launch are stark. More than 96,000 vulnerabilities were published in 2026, more than double the roughly 40,000 of 2024, and two weeks after two critical PHP flaws were disclosed, about 29 million internet-facing services were still running affected versions.
Our Read
Perception answers a problem Microsoft’s own models helped create. The same frontier AI that finds flaws faster than humans can patch them is now being sold back as the defence, and Microsoft is betting agents defending at machine speed are the only match for attackers moving just as fast.
The test is whether security teams will trust agents with real decisions. Microsoft keeps humans in the loop for high-impact actions, but the platform’s value depends on how much autonomy teams actually grant it, and that is a cultural shift as much as a technical one.
The AI Transparency Initiative is the more interesting half of the announcement. If every organisation can see what its AI accessed and did, independently verify those actions and control what is permitted, the industry gets an audit trail for agents, and Microsoft gets to set the standard for it.
Why It Matters
Security work can no longer keep up by hand. With more than 96,000 flaws published this year and exploits appearing within a day of disclosure, a defence that waits for a human to read an alert queue is already late, and Perception is Microsoft’s bid to close that gap.
Background
Perception arrives alongside a wider Ignite security push, and against a live reminder of the stakes. Microsoft has still not explained how an attacker hijacked its Agent Framework packages on PyPI, showing that the supply chain these agents depend on is itself a target.
The platform also sits next to Microsoft’s other Ignite AI moves. Copilot now runs OpenAI’s GPT-6.2 models, and MAI Code 2 brings Microsoft’s own coding model to GitHub Copilot, so Perception is the security layer for an agent stack Microsoft is building fast.
Source: Microsoft Security Blog


0 Comments