Microsoft Purview: Information Protection – Auto-labeling Simulation Evaluation
Microsoft Purview Information Protection will add an AI-powered Simulation Grader feature that evaluates auto-labeling accuracy during policy simulation, showing admins true positive/false positive rates before enforcement. This provides better visibility into policy quality before production rollout.
Key dates
- — preview (Feature currently in development)
Microsoft's description
The Simulation Grader helps admins understand and improve auto‑labeling accuracy before enforcing a policy. When enabled during simulation, matched items are evaluated by an AI-powered grader to determine whether each match is a true positive or false positive, along with an explanation of why the content matched. The results roll up into clear policy‑level accuracy insights and file‑level evaluation details, giving admins confidence to refine conditions, upgrade to smarter classifiers, and move policies to enforcement with greater trust.