When you use Copilot in Microsoft 365 apps, the content it generates can contain confidential business data. Without protection, that output might be shared or stored without the proper security controls. Microsoft Purview sensitivity labels let you automatically apply classification and encryption to files and emails created by Copilot. This article explains how to configure Purview auto-labeling policies so that Copilot outputs inherit the correct sensitivity label based on the data they contain.
Key Takeaways: Configuring Sensitivity Labels for Copilot Outputs
- Microsoft Purview compliance portal > Information protection > Auto-labeling: Create policies that scan Copilot outputs for sensitive content and apply labels automatically.
- Sensitive info types and trainable classifiers: Define which data patterns trigger a label, such as credit card numbers or custom confidential terms.
- Simulation mode before enforcement: Test auto-labeling policies on Copilot outputs without affecting user access to verify correct label assignment.
How Purview Sensitivity Labels Work with Copilot
Microsoft Purview sensitivity labels classify and protect data based on its sensitivity. When Copilot generates content in Word, Excel, PowerPoint, or Outlook, the resulting file or email can be automatically labeled if a matching auto-labeling policy exists. The policy scans the output for sensitive information types, such as Social Security numbers, financial account numbers, or custom patterns you define. If a match is found, the label is applied at the time of creation or saving, encrypting the file or restricting actions like printing or forwarding as specified by the label configuration.
Auto-labeling policies work for files stored in SharePoint, OneDrive, and Exchange Online. They do not apply to content that remains only in the Copilot chat pane without being saved to a document or email. For Copilot outputs that are saved, the policy evaluates the content against your defined rules. The label is applied client-side when the file is opened in a Microsoft 365 app or server-side when the file is uploaded to SharePoint. This ensures consistent protection even if the file is shared externally.
To use auto-labeling with Copilot outputs, you need a Microsoft 365 E5 or E5 Compliance license. You also need the Purview Information Protection role to create and manage policies. The labels themselves must be published to users or groups that use Copilot. Without published labels, auto-labeling policies cannot apply them.
Steps to Create an Auto-Labeling Policy for Copilot Outputs
- Open the Microsoft Purview compliance portal
Sign in to compliance.microsoft.com with an account that has the Information Protection admin role. In the left navigation, expand Solutions and select Information protection. - Go to Auto-labeling policies
Under Information protection, select Auto-labeling. Click Create auto-labeling policy to start the wizard. - Choose the policy type
Select the template that matches your scenario. For Copilot outputs, choose Custom policy so you can define specific sensitive info types. Give the policy a name, such as “Copilot Output – Confidential Data.” - Select locations for the policy
In the Choose locations step, enable SharePoint sites, OneDrive, and Exchange email. These are the locations where Copilot outputs are saved. For SharePoint, you can restrict the policy to specific site collections or apply it to all sites. - Define the rules for detecting sensitive content
Click New rule. Under Conditions, choose Content contains sensitive info types. Select the info types you want to detect, such as U.S. Social Security Number or Credit Card Number. You can also use trainable classifiers to identify custom confidential patterns. Set the minimum count and confidence level to reduce false positives. - Choose the sensitivity label to apply
Under Action, select Apply a specific label. Choose the label you want, for example, Confidential or Highly Confidential. The label must already be published and configured with encryption settings if needed. - Run the policy in simulation mode
In the Policy mode step, select Simulation. This runs the policy without actually applying labels. After the simulation completes, review the results in the Policy reports section. Check for false positives or missed detections. - Turn on the policy
After testing, edit the policy and change the mode to Enforce. Click Save. The policy now applies labels to new and modified Copilot outputs that match the defined conditions.
Common Issues When Labeling Copilot Outputs
Labels are not applied to Copilot-created files
If auto-labeling does not trigger, verify that the sensitivity label is published to the user who saved the file. Open the Purview compliance portal, go to Information protection > Label policies, and confirm the label policy includes the user or group. Also check that the file location is included in the auto-labeling policy. Files saved to a local drive are not scanned. Only SharePoint, OneDrive, and Exchange locations are supported.
Copilot output contains sensitive data but no label is applied
This usually means the sensitive info type condition is not met. Review the simulation report to see which items matched. If the data pattern you expect is not detected, you may need to adjust the minimum count or confidence level. For custom data, create a sensitive info type in Information protection > Sensitive info types before configuring the auto-labeling rule.
Encrypted Copilot outputs cannot be opened by other users
The sensitivity label may include encryption that restricts access. Check the label settings under Information protection > Labels > Edit label > Encryption. If the label is configured with user-defined permissions, only specific users or groups can open the file. To allow broader access, use a label with predefined permissions or assign access to a security group that includes all intended users.
Auto-Labeling Policy vs Manual Labeling for Copilot Outputs
| Item | Auto-Labeling Policy | Manual Labeling |
|---|---|---|
| How labels are applied | Automatically based on content scan | User selects label from the sensitivity bar |
| Requires user action | No | Yes, user must choose or confirm the label |
| Best for | High-volume environments where consistency is critical | Ad-hoc labeling or content that varies widely |
| Supported locations | SharePoint, OneDrive, Exchange Online | All Microsoft 365 apps including local files |
| Configuration time | 30-60 minutes initial setup | Minimal, only publish labels |
| Risk of over-classification | Possible if conditions are too broad | Low, user has control |
After you enable auto-labeling, Copilot outputs are protected at the moment of creation. Review the simulation reports monthly to refine your sensitive info types and reduce false positives. For high-security scenarios, combine auto-labeling with a default label policy that applies a baseline label to all Copilot outputs that do not match any rule. This ensures every file has at least a minimum level of protection.