Azure AI Search Adds Purview Sensitivity Labels to Knowledge Sources
Microsoft’s public preview of Purview sensitivity labels in Azure AI Search lets organizations tag sensitive data at the source, which then flows into knowledge bases used by Microsoft Foundry agents and copilots. This means documents containing confidential information can be automatically marked with labels like Confidential or InternalUseOnly during ingestion, ensuring sensitive data is handled with appropriate security measures.
According to Microsoft’s update, labels applied in systems like SharePoint or other data stores propagate through Azure AI Search’s indexing process. This integration allows Foundry-powered applications to respect these labels when querying or generating responses from sensitive information. For example, a copilot might avoid sharing documents flagged with high sensitivity labels unless explicitly authorized.
For organizations handling regulated data (like healthcare or financial records), this feature bridges a critical gap. It automates data classification and enforces privacy policies at scale without manual intervention. However, users should note it is still in preview — expect potential changes or expanded functionality in future versions. The real win here is reduced risk of accidental data exposure while maintaining usability for legitimate users.
- Start testing now: If your environment uses Microsoft Foundry or Azure AI Search, create a sandbox environment to evaluate how labels integrate with your workflows.
- Audit existing data: Before deployment, identify current sensitive data sources to apply labels proactively.
- Train teams: Ensure users understand how sensitivity labels affect data access in AI-powered tools.
Further details about label configurations are available here. Microsoft also documents practical implementation scenarios for developers.
This preview is a strategic move for enterprises prioritizing data sovereignty. By tying sensitivity labels to AI search operations, Microsoft aligns with growing demands for privacy-by-design architectures in enterprise AI solutions.