OK_Computer oklahoma city it
405-252-9691 Book support

Wiki

Copilot Control System — Security and Governance

The three-pillar Copilot Control System framework: security/governance, management controls, and measurement/reporting. Foundational vs. optimized controls mapped to license tiers.

concept · 4 tags

Summary

When you implement Microsoft Copilot and agents, you face new and amplified risks related to security, compliance, privacy, and governance. The Copilot Control System provides a structured framework to mitigate these risks across Microsoft Copilot, Copilot Chat, Microsoft 365 prebuilt agents, and agents created in Copilot Studio published to Microsoft 365 channels.

The Control System has three pillars:

  1. Security and governance — data security, AI security, and compliance/privacy controls.
  2. Management controls — policies and procedures for overseeing the agent lifecycle.
  3. Measurement and reporting — auditing, risk scoring, and executive reporting.

This article covers the security and governance pillar. See Microsoft Copilot prompts, privacy, and data protection for foundational protections, and copilot-studio-governance-and-agent-security-playbook for the agent lifecycle playbook.

Foundational vs. Optimized Controls

The framework distinguishes two tiers of controls based on licensing:

TierScopeRequired licenseTools
FoundationalSecurity and governance in the Microsoft 365 admin center, SharePoint Advanced Management, and Microsoft PurviewA3 / E3 / G3M365 Admin Center, SharePoint Advanced Management, Purview (basic)
OptimizedAdvanced controls in Microsoft Purview and Microsoft Defender for Cloud AppsA5 / E5 / G5Purview DSPM, Defender for Cloud Apps, Insider Risk Management

Data Security

Safeguard your organization’s information. Use Microsoft Purview and SharePoint Advanced Management to assess oversharing risks, get policy recommendations, and take corrective actions. Copilot and agents only access data the user is authorized to see.

Foundational Data Security Controls (A3/E3/G3)

SharePoint Advanced Management & Purview:

  • Identify potentially overshared data across Microsoft 365 using Data access governance reports for SharePoint sites. Send access reviews to owners of overshared sites.
  • Remove organization-wide site access as needed (via SharePoint or PowerShell with restricted access control).
  • Use SharePoint restricted content discovery to limit Copilot/agent access to overshared sites during remediation.
  • Apply site sensitivity labels via Microsoft Purview Information Protection.
  • Identify and archive/delete unneeded content using SharePoint site lifecycle management and Purview Data Lifecycle Management retention policies.

Microsoft Purview (A3/E3/G3):

  • Get notified when new oversharing occurs via Data Loss Prevention.
  • Secure sensitive data through file-level encryption with sensitivity labels.
  • View reports of sensitive data and unprotected files referenced in Copilot/agent interactions via Purview DSPM for AI.
  • Use sensitivity labels to detect sensitive content and manually prompt protection, and to protect files even if moved or downloaded.

Optimized Data Security Controls (A5/E5/G5)

  • Purview DSPM for AI — create data risk assessments targeted to specific M365 locations (SharePoint, OneDrive); receive and act on policy suggestions for oversharing risks.
  • Automated sensitivity labels — detect sensitive content and automatically apply protections.
  • Insider Risk Management — get alerted to risky user actions that deviate from normal behavior; correlate and sequence risk alerts to identify high-severity patterns; apply adaptive protection to escalate strict security policies for at-risk users.
  • Activity Explorer — view prompts, responses, web queries, and referenced files in DSPM for AI.

AI Security

Safeguard AI-powered tools and their data against evolving threats: oversharing, anomalous user behavior, and misuse of generative AI capabilities.

Foundational AI Security Controls (A3/E3/G3)

Copilot includes built-in protections:

  • Blocks prompt injection attacks (jailbreak attempts).
  • Blocks harmful content generation.
  • Detects protected material (copyrighted content).

Microsoft Purview (A3/E3/G3):

Optimized AI Security Controls (A5/E5/G5)

  • Data Loss Prevention for Copilot and agents — prevent processing of certain sensitive files and exclude them from responses.
  • Insider Risk Management — get alerted to risky AI use (prompt injection attempts, sensitive data access).
  • Adaptive protection — block high-risk users from accessing sensitive content via Copilot/agents.
  • Activity Explorer — view prompts, responses, web queries, and referenced files in DSPM for AI.

Compliance and Privacy

Monitor, audit, and manage how Copilot and agent interactions comply with regulatory and internal standards.

Foundational Compliance Controls (A3/E3/G3)

  • Audit — access detailed logs of Copilot/agent interactions via Purview Audit for Copilot.
  • Retention — enforce retention/deletion policies for Copilot/agent interactions and Teams meeting recordings/transcripts via Purview Data Lifecycle Management.
  • eDiscovery — include user prompts/responses in legal holds; search and export content generated by Copilot/agents.

Optimized Compliance Controls (A5/E5/G5)

  • Communication Compliance — alert on possible compliance or ethical violations and start investigations.
  • Compliance Manager — assess and track adherence to regulatory frameworks.

Zero Trust

Zero Trust is a security approach — “never trust, always verify” — built on three principles:

  1. Verify explicitly — authenticate and authorize every request.
  2. Use least-privileged access — grant only the minimum access needed.
  3. Assume breach — continuously validate and monitor.

For Copilot-specific Zero Trust guidance, see Use Zero Trust security to prepare for Copilot.

Sources