Skip to content

Control 2.14: Declarative Agents from SharePoint Governance — Portal Walkthrough

Step-by-step portal configuration for governing declarative Copilot agents created from SharePoint content.

Prerequisites

  • SharePoint Admin role
  • Microsoft 365 Copilot licenses deployed
  • Understanding of declarative agent capabilities and data access
  • Governance framework for agent creation and deployment

Steps

Step 1: Understand Declarative Agent Architecture

Portal: Microsoft 365 Admin Center Path: Admin Center > Agents > Overview

Review how declarative agents work: users can create custom Copilot agents scoped to specific SharePoint sites or document libraries. These agents answer questions based on the defined content scope, making governance of the underlying content critical.

Agent 365 Centralized Governance: Microsoft has introduced Agent 365 as a centralized governance platform for managing agents across the Microsoft 365 ecosystem. Agent 365 consolidates agent inventory, policy management, lifecycle governance, tools, and risk signals into administrative surfaces such as Agents > Overview, All agents > Registry, Settings, and Tools. Organizations should review Agent 365 settings in the M365 Admin Center to establish centralized visibility over deployed agents — including declarative agents, Copilot Studio agents, and third-party agents — and verify which surfaces are licensed and available in the tenant.

Entra Agent ID: Agents can now be assigned an Entra Agent ID for identity management, enabling organizations to treat agents as managed identities within Entra ID. This supports Conditional Access policies, audit attribution, and permission management for agents. Microsoft Learn states Conditional Access for agents requires Microsoft Entra ID P1/P2 plus Microsoft Agent 365 licensing, with Agent 365 license enforcement coming soon. Organizations should verify licensing, enforcement, and agent identity coverage in the tenant before relying on these controls as active evidence.

Step 2: Configure Agent Access, Sharing, and Lifecycle Rules

Portal: Microsoft 365 Admin Center Path: Admin Center > Agents > Settings

The Agents > Settings page exposes several distinct controls — each governs a different aspect of the agent lifecycle and they should not be confused with one another:

  • Allowed agent types — controls which categories of agents (Microsoft, organization-published, external publisher) appear in the agent catalog and can be installed by users. Use this to disallow external-publisher agents until vendor risk assessment is complete.
  • Agent management rules — bulk actions for tenant administrators. The current actions are Install Microsoft agents, Reassign ownerless Agent Builder agents to manager, Block ownerless agents without usage, Apply a template to agents, and Reject publish requests older than a specified number of days.
  • Security templates (policy templates) — reusable security policies for new agents. Do not conflate this with the Apply template management-rule action, which applies a preexisting or custom template to eligible existing published agents.
  • Sharing — controls whether (and which groups of) users can share agents organization-wide. Note: sharing controls apply only to agents built with Microsoft 365 Copilot Agent Builder; SharePoint agents and other agent types are governed through their own publishing surfaces.
  • User access — controls how users access and install agents (All users / No users / Specific users or groups). This setting governs consumption, not creation. Restricting User access to a pilot group is a useful Baseline lever, but it does not restrict who can build a new declarative agent — that is gated by Copilot license, Agent Builder availability, and SharePoint agent permissions on source sites (see Step 3).
  • Agent feedback sharing — controls whether developers receive thumbs-up / thumbs-down feedback and comments. It doesn't change who can rate agents or what an agent can access.
  • Tags — lets administrators manage organization-wide discovery labels. Microsoft currently allows up to 50 tags, and deleting a tag removes it from every agent that uses it.

When reviewing or using Apply template:

  1. Define criteria for the agent instances to evaluate.
  2. Select a preexisting template, or select policies from Custom to create a custom configuration.
  3. Review all eligible agents or choose a subset, and then run the rule.
  4. Record that eligibility is limited to Registry agents that have an agent identity. Agent blueprints and AI teammates are excluded.
  5. Treat the result as a one-time bulk operation. It is not a scheduled rule and does not provide continuous enforcement.

For the current product behavior and limitations, see Agent settings in Microsoft 365 admin center.

For Baseline governance: - Set User access to a small pilot group while the governance framework is being established - Set Allowed agent types to Microsoft and organization-published only - Disable broad Sharing for Agent Builder agents until an approval workflow exists

Step 3: Review Agent Data Access Scope

Portal: SharePoint admin center Path: SharePoint Admin > Active Sites > review sites used as agent data sources

For each declarative agent, the data access scope is defined by the SharePoint site or library it references. Verify: - Sites used as data sources have appropriate sensitivity labels and permissions - Content is appropriate for the agent's intended audience - No oversharing exists on the source site (cross-reference with Control 1.2)

Step 4: Configure Agent Sharing and Distribution

Portal: Microsoft 365 Admin Center Path: Admin Center > Agents > Settings > Sharing

Control how declarative agents are shared: - Limit broad sharing to approved groups or disable it where appropriate; the Agent Builder Sharing control currently offers All users, No users, or Specific users - Disable organization-wide agent publishing for unapproved agents - Require governance review before broad distribution - Separately verify that the underlying SharePoint site, library, or file permissions still enforce the intended content boundary for anyone who can access the agent

Step 5: Document Agent Governance Framework

Create the agent governance document covering: - Who can create agents and under what conditions - Data access review requirements for agent source content - Testing requirements before agent publication - Ongoing monitoring and review of active agents - Decommissioning process for obsolete agents - Third-party model provider policy — document the organization's policy on third-party model providers (e.g., non-Microsoft AI models accessed through agents). Verify whether the tenant exposes a third-party model-provider setting or equivalent control before treating it as configured. If the setting is present and disabled, document that state; if enabled, require a specific business case, completed vendor risk assessment, and documented data-residency implications. If the setting is not present in the tenant, record that absence rather than assuming a disabled default.

FSI Recommendations

Tier Recommendation
Baseline Restrict agent access/publishing to approved groups where the tenant exposes controls; review data sources before agent deployment; verify any third-party model-provider setting in the tenant
Recommended Formal agent approval process; source site security verification; agent inventory tracking via Agent 365; tenant-verified Entra Agent ID coverage where applicable; documented third-party model-provider policy
Regulated Broad publication requires governance committee approval; data impact assessment for each agent; quarterly agent review; comprehensive audit logging; Entra Agent ID and Conditional Access coverage verified where licensed and applicable; third-party model providers disabled or restricted to approved providers with completed vendor risk assessment

Next Steps