Sensitivity Label Recommendations

info

Machine learning feature - no LLM involved, no AI toggle required. Analysis runs entirely within Syskit Point. Take a look at the AI Data Privacy & Security article for more details.

Unlabeled or wrongly labeled workspaces are open exposure windows; the policies and protections tied to sensitivity labels simply don't apply to them. And manually classifying every new and legacy workspace takes up time.

Sensitivity Label Recommendations help with that. Machine learning analyzes the workspaces your organization has already labeled, learns what each label looks like in practice, and:

• Recommends labels for unlabeled workspaces.
• Flags existing labels that look wrong.
  • For example, a workspace labeled General that behaves exactly like your Confidential workspaces deserves a second look.

Suggested Sensitivity Labels Report

The Suggested Sensitivity Labels Report functions in the following way:

• The model learns from your organization's own labeling decisions - membership patterns, content signals, sharing behavior, and workspace characteristics of already-labeled workspaces.
• It scores unlabeled workspaces against those patterns and proposes the best-fitting label.
• Recommendations appear in the Suggested Label column.

For more details on the report itself, please take a look at the Sensitivity Labels Reports article.

info

Please Note that suggested sensitivity labels:

• Are admin-confirmed and never auto-applied.
  • No label is ever applied without an explicit admin decision.
• Improve with coverage.
  • The more workspaces you label (and the more recommendations you confirm or correct), the better the recommendations get.
• Provide faster classification of new workspaces.
  • Newly created and unmanaged workspaces get a recommendation sooner, lessening the time they remain unlabeled and potentially unprotected.

Prerequisites

• Sensitivity labels are set up in your Microsoft 365 tenant and visible to Syskit Point. Take a look at the Enable Sensitivity Labels article.
• There is a meaningful base of already-labeled workspaces to learn from.