Privacy Considerations and PII Protection with Copilot

Overview

Federal agencies must comply with the Privacy Act and protect personally identifiable information. This video examines the privacy considerations specific to M365 Copilot, including PII handling, Privacy Act compliance, and technical controls to minimize privacy risks.

Essential for privacy officers, legal counsel, security teams, and agency leadership responsible for privacy compliance.

What You’ll Learn

• Privacy Act Basics: How the Privacy Act applies to AI-enabled systems
• PII in Copilot: How Copilot accesses and processes personally identifiable information
• Privacy Impact Assessment: Conducting PIAs for Copilot deployments
• Technical Controls: DLP, access controls, and monitoring for PII protection
• User Education: Training employees on privacy-aware Copilot use

Transcript

[00:00 - Introduction]

Welcome everyone. Kevin Tupper here with Sarah Johnson. Today’s topic is privacy—specifically, how to deploy M365 Copilot while protecting personally identifiable information and complying with the Privacy Act and other federal privacy requirements.

[00:45 - Why Privacy Matters for AI]

Copilot’s power comes from accessing your organization’s data—emails, documents, chat history. Inevitably, some of that data contains PII: employee records, constituent information, case files, and more.

The Privacy Act requires agencies to:

Collect only PII that’s relevant and necessary. Maintain accurate and complete records. Establish appropriate safeguards. Limit disclosure of PII without consent. Provide individuals access to their own records.

Copilot must be implemented in a way that respects these principles.

[02:30 - How Copilot Handles PII]

First, let’s clarify what Copilot does and doesn’t do with data:

Copilot does NOT store your prompts or responses for training Microsoft’s AI models. Your data stays in your tenant. Copilot does NOT create new copies of PII. It references existing M365 content. Copilot DOES respect existing permissions. Users can only access PII via Copilot if they already have permission to access that data directly. Copilot logs interactions to the audit log for monitoring and compliance purposes.

So Copilot itself doesn’t introduce new PII handling—it amplifies access to existing PII.

[04:30 - Privacy Impact Assessment]

Before deploying Copilot, conduct a Privacy Impact Assessment. Your PIA should address:

What PII will Copilot potentially access? Employee data, constituent records, case files, personnel information?

What is the purpose of Copilot accessing this PII? To assist with case management, HR workflows, policy drafting?

Are there alternatives that involve less PII exposure? Could you restrict Copilot to certain departments or exclude certain data repositories?

What safeguards will protect PII? Access controls, DLP policies, audit logging, encryption?

What are the risks if PII is mishandled? Impact to individuals, agency, mission?

Your Senior Agency Official for Privacy (SAOP) should review and approve the PIA before deployment.

[06:45 - Technical Safeguards]

Implement layered technical controls:

Access controls: Ensure only authorized users have Copilot licenses and access to PII repositories. Use conditional access to restrict Copilot usage to compliant devices and approved locations.

DLP policies: Configure Data Loss Prevention to detect and block PII in Copilot-generated content. For example, prevent Copilot from suggesting email content that includes SSNs or sensitive employee information.

Sensitivity labels: Apply labels to documents containing PII, and configure policies that restrict how those documents can be used with Copilot.

Audit logging: Enable comprehensive logging of all Copilot interactions with PII. Review logs regularly for anomalies or policy violations.

[08:30 - Data Minimization]

Privacy by design means collecting and processing only the minimum PII necessary. For Copilot:

Evaluate whether all data repositories need to be accessible via Copilot. Could you exclude HR systems or case management databases containing highly sensitive PII?

Configure Microsoft Graph to limit what data Copilot can index. Use exclusion policies to keep certain SharePoint sites or mailboxes out of Copilot’s reach.

Train users on privacy-aware prompting. Encourage users to be specific in their prompts rather than asking Copilot to “search all files about person X,” which might expose unnecessary PII.

[09:45 - User Education]

Technology alone isn’t enough. Educate users on privacy responsibilities:

Don’t use Copilot to aggregate or analyze PII without legitimate need-to-know. Verify accuracy of Copilot outputs before using them in decisions affecting individuals. Don’t share Copilot-generated content containing PII outside authorized channels. Report privacy concerns or incidents immediately.

Include privacy training as part of your Copilot onboarding program.

[10:30 - Ongoing Compliance]

Privacy compliance isn’t one-time—it’s continuous:

Quarterly reviews of PII access patterns via audit logs. Annual PIA updates as Copilot usage evolves. Regular consultation with your SAOP on emerging privacy risks. Integration of Copilot into broader agency privacy management program.

[11:05 - Conclusion]

M365 Copilot can be deployed in a privacy-compliant manner with proper planning, technical safeguards, and user education. By conducting a thorough PIA, implementing layered controls, and training users on privacy principles, you protect individuals’ rights while enabling AI productivity. Download our Privacy Impact Assessment Template linked below to guide your evaluation.