The Agent Proliferation Challenge in Government IT
Artificial intelligence agents are no longer a future consideration for government IT leaders - they are here, and they are multiplying fast. From AI assistants built inside Microsoft Copilot Studio to agents wired into HR workflows, permit processing, or constituent service portals, public sector organizations are discovering that deployment is the easy part - governance is the challenge.
Who built that agent? What data can it access? Is it still being used? Who owns it if the original creator leaves the organization? These are not hypothetical questions - they are active compliance and security risks for agencies operating in complex regulatory environments.
Microsoft has been investing heavily in answering these questions with a centralized governance layer for AI agents, built directly into the Microsoft 365 admin center. This post walks through what those tools look like today, what is on the roadmap, and why this matters especially for government organizations operating inside GCC or hybrid Microsoft 365 environments.
What Is the Agent Management Experience in Microsoft 365 Admin Center?
The Microsoft 365 admin center now includes a dedicated Agents section that serves as the control plane for every AI agent operating inside your Microsoft 365 tenant. Administrators with the AI Admin role (or Global Reader for view-only access) can access this panel and take action across several key areas.
The Agent Registry
The Agent Registry is a centralized inventory of every agent integrated with Microsoft 365 Copilot in your organization. It surfaces four categories of agents:
- Microsoft agents - built and maintained directly by Microsoft
- External partner-built agents - created by trusted non-Microsoft developers in the Teams app ecosystem
- Shared by creator - agents built and shared by individuals inside your own tenant, typically using Copilot Agent Builder or Copilot Studio
- Published by your organization - custom agents that have gone through an admin approval and publishing workflow for broader organizational use
For each agent, administrators can view metadata including the agent’s name, publisher, creation date, host products, and availability status. The registry also surfaces which agents use embedded file content as a knowledge source - and what sensitivity labels have been applied to those files via Microsoft Purview - providing the audit trail needed to demonstrate data governance compliance.
One important confirmation for government tenants: publishing agents to the organization is supported in both Government Community Cloud High (GCCH) and Government Community Cloud Moderate (GCCM) environments, making the approval and publishing workflow available to M365 GCC customers today.
Agent Settings: Your Policy Enforcement Layer
The Agent Settings page provides four categories of org-wide governance controls:
Allowed Agent Types lets administrators restrict which categories of agents users can discover and install from the agent catalog. You can independently toggle on or off Microsoft-built agents, your own organization’s custom agents, and external partner agents.
Sharing Controls define who can share agents inside your organization. Options range from allowing all users to share with anyone in the org, to limiting sharing to designated groups, to disabling org-wide sharing (note: users retain the ability to share directly with specific individuals even when org-wide sharing is off). Only agents built with Agent Builder are currently governed by this setting.
Templates allow administrators to apply a preset bundle of governance policies when an agent is activated or published. Default templates pull in security and compliance controls from Microsoft Entra, Microsoft Purview, and SharePoint automatically. Custom templates allow layering additional controls - such as restricting access to external content - on top of those defaults.
User Access is the final lever. Administrators can restrict agent access to all users, no users, or a specific set of users and groups - critical for phased rollouts or pilot programs.
The Agent 365 Overview Dashboard: Visibility at Scale
Beyond per-agent management, Microsoft has introduced the Agent 365 Overview page as a governance command center for the entire tenant. It surfaces hero metrics for the past 30 days:
- Agent Registry count - total agents in your catalog
- Active users - unique users who interacted with at least one agent
- Total sessions - complete agent invocations across the tenant
- Exception rate - percentage of sessions that completed without errors
- Agent Runtime - total AI-assisted time across all sessions
The dashboard highlights actionable governance gaps: pending agent approval requests, agents without assigned owners, and exceptions needing administrator review. An Agents by Publishers and Agents by Platform breakdown shows which creation platforms (Copilot Studio, Azure AI Foundry, external partners) are generating the most agents in your environment.
Note: The real-time usage metrics (active users, sessions, exception rate, agent runtime) are currently available to organizations enrolled in the Frontier preview program with a Microsoft 365 Copilot license. The core agent registry and settings capabilities are available to all M365 Copilot tenants today.
Microsoft Entra Agent ID: Identity for AI Agents (Preview)
Looking ahead, Microsoft is extending its identity platform to AI agents themselves through Microsoft Entra Agent ID. This is a significant architectural shift: instead of agents operating as anonymous or system-level processes, each agent gets its own managed identity in Microsoft Entra.
With Entra Agent ID (currently in Frontier preview), IT teams gain:
- Conditional Access for agents - enforce Zero Trust policies at the agent identity level, not just the user level
- Risky agent detection - Microsoft Entra ID Protection detects and blocks threats by flagging anomalous activities and unusual or unauthorized actions involving agents, similar to how it monitors risky users
- Agent lifecycle workflows - govern agent identities from deployment to expiration, ensuring sponsors and owners are assigned and access is intentional, auditable, and time-bound
- Audit logs - activity logs for agents, integrated with your existing SIEM and Microsoft Sentinel workflows
This mirrors how government organizations already govern service accounts and non-human identities - purpose-built for AI agents.
Why This Matters for Government
Government organizations operate under a unique set of pressures that make AI agent governance not just a best practice but a legal and operational necessity.
Compliance and auditability are non-negotiable in public sector IT. Whether it is CJIS, IRS 1075, NIST 800-53, or state-level data classification requirements, government agencies need to demonstrate that AI tools are operating within defined boundaries, accessing only approved data, and that there is a clear chain of ownership and accountability for every automated system. The Agent Registry, sensitivity label integration, and audit log capabilities directly support these requirements.
Shadow AI is a real risk. When employees build and share agents without IT visibility, organizations lose control of what data those agents can access. The Agent Settings page gives administrators the tools to prevent ungoverned agent sprawl before it becomes a security incident.
Workforce transitions create orphaned agents. Government agencies regularly deal with staff turnover, retirements, and reorganizations. An agent built by an employee who has since left - and still has access to sensitive SharePoint sites or HR data - is a liability. The ownerless agents alert in the Agent 365 overview dashboard makes this a visible, actionable problem rather than a hidden one.
Zero Trust alignment is a federal mandate that is cascading through state and local government as well. Extending Conditional Access and least-privilege principles to AI agents - rather than carving out a governance exception for them - keeps government IT environments consistent with their existing security architecture.
For organizations already operating in M365 GCC or GCCH, the core agent governance capabilities in the Microsoft 365 admin center are available today. The more advanced Agent 365 Frontier preview features (real-time metrics, Entra Agent ID) are currently rolling out in commercial tenants first, so GCC customers should watch the Microsoft 365 Government feature roadmap for availability timelines.
Getting Started
If you have a Microsoft 365 Copilot license and the AI Admin role in your tenant, the Agents section of the Microsoft 365 admin center is available to you today. A practical starting point:
- Audit your current agent inventory using the Agent Registry - you may be surprised how many agents already exist in your tenant
- Review and configure Agent Settings to align with your organization’s acceptable use and data governance policies
- Identify ownerless agents and assign appropriate ownership before a governance gap becomes an incident
- Establish an agent approval workflow using the publishing process, so new agents go through a review before reaching broad audiences
- Explore the Frontier preview program if your organization wants early access to the full Agent 365 dashboard and Entra identity capabilities
For government IT leaders, Microsoft’s investment in centralized agent governance makes one principle clear: adopting AI at scale should never mean sacrificing control.
Additional Resources