Purview can block sensitive data in AI prompts: turn it on before Copilot rollout
🚨 The Signal: Microsoft Purview can block sensitive data -- including credit card numbers and personal data -- from being pasted into Microsoft 365 Copilot and third-party AI apps. Endpoint DLP enforces this at the browser. Without it, staff can leak regulated data into AI prompts in seconds.
The Impact
Generative AI amplifies oversharing; Purview controls let you warn or block sensitive data entering prompts and surface where it already happens.
- Purview / compliance admins: configure DLP for AI before broad Copilot rollout.
- End users: pasting regulated data into AI tools can be warned or hard-blocked.
- Data owners: oversharing risk amplified by AI surfacing labelled content.
- Risk: PII and financial data leaking into third-party LLMs (ChatGPT, Gemini, DeepSeek).
The Action
- In the Microsoft Purview portal, open Data Security Posture Management (DSPM) for AI to discover AI usage.
- Onboard Windows devices to Purview, then create an Endpoint DLP policy that blocks sensitive info types on generative-AI sites.
- Enable sensitivity labels for SharePoint and OneDrive so Copilot honours the EXTRACT usage right.
- Switch on the Insider Risk Management Risky AI usage policy template to detect prompt injection and protected-material access.
- Enable Audit and Communication Compliance for Copilot prompts and responses.
Domain: Purview · Impact: high · Workload: Microsoft Purview