Microsoft Purview: Data Security Triage Agent in Data Loss Prevention - Reasoning Trace and Confidence Score for agent-triaged alerts

This feature introduces confidence scoring and reasoning trace explainability into the Data Security Triage Agent in DLP, addressing a critical trust gap reported across multiple customers. Today, without transparency into why the agent makes a decision or how confident it is, analysts are forced to fall back to manual review — completely negating the automation value. The feature will surface a reasoning trace alongside each agent decision and a confidence score, giving analysts a clear signal of how reliable the agent's output is. This provides SOC teams with an auditable, explainable output they can validate rather than blindly trust, enabling them to progressively increase reliance on automation over time.

Affected: Microsoft Purview

Related changes

• Microsoft Purview: Data Loss Prevention - Data Security Triage Agent summaries and categorizations for DLP alerts will be made available in DLP Alerts in Microsoft Defender XDR · Data
• Microsoft Purview: Data Loss Prevention - User Based Aggregation of DLP Alerts · Data
• Microsoft Purview: Data Loss Prevention - Data Security for non-Microsoft connected apps · Data
• Microsoft Purview: Data Loss Prevention - Rule-based tagging and auto-resolution of alerts based on business needs of the customer. · Data
• Microsoft Purview: Data Loss Prevention - Data Loss Prevention to safeguard sensitive information from external web search in Microsoft 356 Copilot and Copilot Chat · Data
• Microsoft Purview compliance portal: Data Loss Prevention - DLP Policy Tip Support in Outlook for Mac · Data

Change history

View at Microsoft → Copy summary