Copilot Sensitive Info type DLP safeguarding not blocking prompts as intended

The behavior matches the current design and limitations of DLP for Microsoft 365 Copilot prompts rather than a rollout or licensing issue.

Key points from the documented behavior:

1. The “block sensitive information types in prompts” feature is in preview and only applies to the Microsoft 365 Copilot and Copilot Chat policy location.
2. DLP evaluates only the text typed directly into the Copilot prompt. It does not inspect:
   • Files uploaded into the prompt
   • Other content sources used by Copilot (SharePoint, OneDrive, Exchange, etc.)
3. When the Content contains → Sensitive information types condition is used with the action Prevent Copilot from processing content → Processing prompts, the expected result is:
   • Copilot does not respond to the prompt.
   • The prompt is not used for internal or external web searches.
4. There is a separate action, Prevent Copilot from processing content → Performing Web Searches, which only blocks external web search but still allows Copilot to answer using internal data.

Given this, the DLP rule must meet all of the following to fully block the prompt:

1. Policy location is Microsoft 365 Copilot and Copilot Chat (no other locations in the same policy).
2. Condition is Content contains → Sensitive information types.
3. Action is specifically Prevent Copilot from processing content → Processing prompts (not just “Performing Web Searches”).
4. The policy is in enforce mode (not simulation) and has had time to sync (up to four hours) into Copilot.

If Copilot is:

• Flagging content as sensitive,
• Continuing to respond,
• And no DLP alerts appear in Purview,

then the most likely configuration issue is that the rule is using the web search–only action or a different location/condition combination, so the prompt itself is never being blocked by the Copilot-specific DLP engine.

Recommended checks and adjustments:

1. Open the DLP policy and confirm:
   • Location: Microsoft 365 Copilot and Copilot Chat is selected.
   • No other locations are enabled in that same policy (this location disables others by design).
2. Edit the rule and verify:
   • Condition: Content contains → Sensitive information types with the SITs being tested.
   • Action: Prevent Copilot from processing content → Processing prompts is selected.
3. Ensure the policy is Turned on immediately (not simulation) and wait up to four hours for changes to apply.
4. Test again using only typed prompt text that matches one of the configured SITs.

If all of the above are confirmed and the prompt still processes with no DLP event, the remaining path is to open a support case so the service team can check for tenant-specific issues with the preview rollout of the Microsoft 365 Copilot and Copilot Chat policy location.

References:

• Learn about using Microsoft Purview Data Loss Prevention to protect interactions with Microsoft 365 Copilot and Copilot Chat
• Use Microsoft Purview to manage data security & compliance for Microsoft 365 Copilot & Microsoft 365 Copilot Chat
• Step 3: Block sensitive data going to sanctioned AI apps
• Learn about the default data loss prevention policy for Microsoft 365 Copilot location