Skip to content

Control 2.14: Declarative Agents from SharePoint Governance — Troubleshooting

Common issues and resolution steps for declarative, SharePoint-backed, and Registry-governed agent security.

Common Issues

Issue 1: Users Creating Agents Without Governance Approval

  • Symptoms: Declarative agents discovered in the tenant without governance approval or documentation
  • Root Cause: Agent access, sharing, publication, or the specific authoring surface may not be properly governed. User access limits consumption and installation; it does not by itself prove that every creation path is blocked.
  • Resolution:
  • Review agent settings in Admin Center > Agents > Settings, including Allowed agent types, Sharing, User access, and management rules.
  • Review the specific authoring surface used (Agent Builder, SharePoint agent creation, Copilot Studio, or pro-code package upload) and apply the control that surface supports.
  • Restrict broad access, installation, sharing, and publication to approved security groups where the tenant exposes controls.
  • Audit existing agents and require retroactive governance approval.
  • Set up monitoring (Script 3) to detect new agent creation or publication.

Issue 2: Agent Accessing Overshared Source Content

  • Symptoms: A declarative agent returns sensitive content that the querying user should not have access to
  • Root Cause: The agent's source SharePoint site has oversharing — content is accessible to more users than intended.
  • Resolution:
  • Immediately restrict the source site sharing settings
  • Run an oversharing assessment (Control 1.2) on the source site
  • Remediate permissions before re-enabling the agent
  • Consider suspending the agent until remediation is complete

Issue 3: Agent Providing Inaccurate or Stale Responses

  • Symptoms: Agent responses reference outdated content or provide incorrect information
  • Root Cause: Source content may be outdated, or the semantic index may not have processed recent updates to the source site.
  • Resolution:
  • Verify source content is current and accurately maintained
  • Request a re-index of the source site if recent updates are not reflected
  • Add content freshness indicators to the agent description
  • Establish a content review cadence for agent source sites

Issue 4: Agent Governance Process Slowing Deployment

  • Symptoms: Business teams report agent approval takes too long
  • Root Cause: Governance process may be overly complex for low-risk agents.
  • Resolution:
  • Create tiered governance based on data sensitivity and audience scope
  • Fast-track agents referencing already-approved, properly governed sites
  • Pre-approve common agent patterns with standardized templates
  • Define clear SLAs for governance review

Issue 5: Third-Party Model Provider Enabled or Unverified

  • Symptoms: Agents are using non-Microsoft AI model providers to process organizational data without the compliance team's knowledge, or the tenant's third-party model-provider posture has not been verified.
  • Root Cause: A tenant-exposed third-party model provider setting or equivalent control was enabled without governance review, or the tenant does not expose a clear setting and the posture was assumed rather than evidenced.
  • Resolution:
  • Review the current M365 Admin Center / Copilot / Agent settings surfaces for any third-party model-provider setting or equivalent control.
  • If enabled without governance approval, disable it immediately and notify the compliance team.
  • If no setting is present, record "not available in tenant" with date and reviewer; do not treat absence as a disabled default.
  • Audit whether any agents used third-party model providers while the setting was enabled — review audit logs for non-Microsoft model invocations where available.
  • Complete a vendor risk assessment for any third-party model provider before re-enabling.
  • Document the approved third-party model provider policy and communicate to agent creators.

Diagnostic Steps

  1. Check agent inventory: Review Admin Center > Agents > All agents / Registry (or Agent 365 dashboard)
  2. Verify source security: Run Script 2 on agent data sources
  3. Review creation / publishing policies: Verify the specific controls for each authoring surface rather than inferring creation restrictions from User access
  4. Check third-party model providers: Verify whether a setting exists, and if present whether it is disabled or restricted to approved providers
  5. Monitor activity: Run Script 3 for recent agent events
  6. Test agent scope: Query the agent to verify content boundaries

Escalation

Severity Condition Escalation Path
Low Governance process improvement needed Governance team
Medium Unauthorized agent creation detected Security Operations for review
High Agent exposing sensitive content Security Operations and site owner
Critical Regulated data exposed through ungoverned agent CISO and Compliance Officer immediately