Library - Product reference
Articles open in this browser tab. Some require sign-in.
Blocking an AI agent contains the immediate threat. It does not complete the response. Once an alert is confirmed and the agent or affected instance is blocked, the security team needs to answer a harder question: what did the agent do before we stopped it?
Sep 16, 2026
Your organization has one agent estate. So why do the Microsoft portals show different totals? The gap can look like shadow AI, a missing identity or a broken integration. Sometimes, you are simply comparing different populations.
Sep 08, 2026
The riskiest part of an MCP integration is not the connection. It is the authority that crosses it! An approval should describe authority, not only connectivity. The effective authority of an MCP integration is created by six things.
Aug 10, 2026
Identity provides traceability. Governance provides answers. An agent can have a valid security principal and still lack a complete governance record.
Aug 06, 2026
An agent incident can involve nine teams and still leave one critical question unanswered: "Who is allowed to decide what happens next?" The problem is not always missing technology. It is often that actors, products, evidence, actions, and decisions are drawn as if they mean the same thing.
Aug 04, 2026
"Does the agent work?" - Most agent governance problems begin with this wrong production question! While a successful demo answers that question, a production approval has to answer a different one: "Can this agent operate with bounded authority, observable behavior, and a clear human decision owner?"
Jul 21, 2026
Be careful with some of the shorthand we use. When we say organizations need visibility, permissions, and traceability for agents, that can easily sound as if no controls exist today. That is not accurate, and any average security team will challenge it immediately.
Jun 25, 2026
CTOs, CISOs, and AI leaders: try this with your teams. Pick the different agent examples shown in the visual and ask one question: Would we govern, secure, and control each of these the same way
May 27, 2026
As organizations move from isolated AI experiments to a mixed agent ecosystem, the hard part is no longer only creating agents. The hard part is seeing them clearly, understanding where they came from, knowing who is accountable, controlling what they can reach, and connecting their activity to security and compliance signals.
May 25, 2026
The workplace is no longer only about people using apps and accessing data. It is becoming AI-powered, Copilot-assisted, and increasingly agent-operated. That shift creates a different governance question: “Can this user, Copilot, or agent access this data, use this tool, take this action, and produce evidence that it stayed within policy?” That is where Microsoft 365 E7 becomes important.
May 19, 2026
