Skip to content
AI agents are your newest insiders: a Microsoft 365 checklist

AI agents are your newest insiders: a Microsoft 365 checklist

· Insights

Microsoft is treating AI agents as a new kind of insider: software that holds its own identity, reaches company data, and acts without someone clicking each step. On October 19, 2026, Microsoft published its security approach for them, and its Microsoft 365 roadmap lists related controls with disclosure or preview dates from October 2026 to January 2027. If your staff use Copilot, Copilot Studio or any coding or browser agent, this affects you even if you never planned to deploy agents.

What Microsoft announced

In a post dated October 19, 2026, Nadim Abdo, Corporate Vice President for Identity and Network Access at Microsoft, describes how Project Perception extends to what Microsoft calls the agentic insider. The post frames the job as four responsibilities: know which agents exist, contain what they can access, protect the resources in their path, and adapt protection as risk changes.

The pieces Microsoft names include:

  • Agent discovery. Coverage grows to 60 supported local agents, including Claude Code, OpenAI Codex and GitHub Copilot, with signals shared with Microsoft Agent 365.
  • Microsoft Entra Agent ID. Each agent gets its own identity, separate from the person who uses it, with identity types for more agent patterns.
  • Dynamic Permissions. Agents start with no authority and receive task-scoped, time-bound permissions.
  • Dynamic Agent Risk and Dynamic Conditional Access. Risk is assessed continuously, and the response is graduated, from reducing permissions up to stopping execution.
  • Microsoft Execution Container. A sandbox that runs agents under policy-controlled access.

The post does not give availability dates or licensing for these capabilities, so do not assume they are included in your current plan. Microsoft points to its Ignite conference, November 17-20, 2026, as the next place to hear more.

Roadmap items that are dated

Separate from the blog post, three Microsoft 365 roadmap entries carry dates. None of them lists a general availability date:

  • Agent 365 reporting in Viva Insights (Copilot Analytics). Microsoft lists public disclosure for October 2026, with the status in development. Customers can view agent insights in the Agent 365 Dashboard without an Agent 365 license, with reporting on Microsoft platform-built agents and context on who built each agent and why.
  • Copilot Studio support for Microsoft Purview Information Protection sensitivity labels. Microsoft lists public disclosure and preview for October 2026. Users can upload labeled files to Copilot Studio agents deployed on the Microsoft 365 channel, and agents process content based on the user’s and agent’s permissions, with label protections enforced end to end.
  • Work IQ: Agent Policy Governance. Microsoft lists preview for November 2026 and public disclosure for January 2027. It gives administrators policy-based controls for governing agents and their capabilities at scale. The roadmap entry is short, so details of the controls are not yet public.

Roadmap dates can move, and a disclosure or preview month is not a release date. Treat the months above as Microsoft’s current plan, not a commitment. The entries appear in the Microsoft 365 roadmap, and we confirmed their date fields through Microsoft’s release communications API.

Why it matters for a small business

An agent that acts for a user inherits that user’s reach. If an employee can open every folder in SharePoint because permissions were never tidied, an agent working for them can read all of it, quickly, and may surface it in an answer to the wrong person. Microsoft’s own framing of least privilege, with task-scoped and time-bound access, is the principle small offices have long been told to follow for people. Agents raise the cost of ignoring it.

Sensitivity labels matter for the same reason. For defense contractors that handle controlled unclassified information, a label that follows a file into an agent is far better than a policy that exists only in a handbook.

What to do

  1. List the agents already in use. Ask each department what they have turned on: Copilot, Copilot Studio agents, browser or coding agents, and any third-party tools that connect to Microsoft 365. You cannot govern what you have not counted.
  2. Decide who may create agents. In the Microsoft 365 and Power Platform admin centers, check whether every licensed user can build and share an agent. Limit it to named people until you have a policy.
  3. Review who can reach sensitive folders. Remove broad groups such as Everyone except external users from finance, HR and contract libraries before an agent makes that access useful to someone who should not have it.
  4. Apply sensitivity labels to your most sensitive documents. Start with a small set, such as payroll, client contracts and, if relevant, CUI. Labels are the control the Copilot Studio preview builds on.
  5. Look at the Agent 365 view once it appears. When the Viva Insights dashboard is available in your tenant, record who built each agent and whether anyone can explain its purpose.
  6. Write a one-page rule for staff. Say which agents are approved, what data must never be pasted or connected, and who to ask. Keep it short enough that people read it.
  7. Keep Conditional Access and multifactor sign-in healthy. Agent controls sit on top of identity. If sign-in protection for people is weak, agent protection will not make up for it.

What can wait

You do not need to buy anything or redesign your tenant because of this announcement. Dynamic Permissions, the Microsoft Execution Container and Work IQ policy governance have no confirmed licensing in the sources we reviewed, and the blog post gives no availability dates for them. Wait for details after Ignite on November 17-20, 2026, and for Microsoft’s January 2027 disclosure date for Agent Policy Governance, before planning around them. The inventory, permission cleanup and labeling above cost little and help regardless of what ships.

Where we can help

Summit Networks helps Front Range businesses review Microsoft 365 permissions and set sensible rules for AI tools; see our AI integration work or contact us to talk through your setup.

Sources

TaggedAI agentsCopilotEntragovernanceMicrosoft 365

← All insights