Ask most IT admins how many Copilot agents are deployed in their Microsoft 365 tenant right now. Almost none can answer. Not because they haven't looked — it's because until recently there was no easy way to look. The agents accumulated quietly: a developer built one in Copilot Studio, a department head installed a third-party agent from the Teams store, Microsoft rolled out built-in agents with the E5 licence renewal.
This matters because agents are fundamentally different from the Copilot chat experience most conversations focus on. Interactive Copilot waits for the user to ask a question. Agents act autonomously — they read files, send messages, create calendar entries, and take actions based on triggers, schedules, or instructions — as the user, on the user's behalf, without the user initiating each action.
Agents operate using the identity of the user who activates or interacts with them — they inherit that user's permissions. If that user can read the Finance SharePoint site, the agent can read it. If that user can send email, the agent can send email. This means your existing governance controls — DLP policies, sensitivity labels, Conditional Access, Insider Risk — become the primary controls against agent misuse. Without them extended to cover Copilot interactions, agents can access and surface data that DLP would otherwise restrict.
The broad deployment problem: An agent deployed to "all users" is available to every person in your organisation. If that agent has write access to SharePoint or can send email on behalf of users, you've effectively given a third-party system the ability to take actions across your entire tenant. Most organisations that have deployed agents broadly haven't thought through this explicitly.
The Microsoft 365 admin centre has an Agent management section under Settings → Copilot → Agent 365. This shows agents installed or published in your tenant, their publication scope, and blocked/active status. For each agent you can see who published it, which hosts it's available in (Copilot, Teams, Outlook), and whether it's available to all users or specific groups.
Before any agent — custom or third-party — is made available to all users: DLP policies covering Copilot and agent interactions, sensitivity labels published and applied to sensitive content, Conditional Access enforced with MFA required, Purview audit logging active and capturing Copilot interactions, and a review of what data the agent can access and whether that access is proportionate to its purpose.
API availability note: The Copilot Agent Catalog API is in Public Preview and rolling out tenant-by-tenant. If your tenant hasn't received the rollout yet, M365Clarity will show this check as Not Verified rather than reporting a false result.
M365Clarity inventories all Copilot agents and flags any deployed broadly without governance controls in place.
Run a free scan →Related articles