Sensitivity Labelling
also called Data Classification Tagging
Attaching a machine-readable classification to data at the column or asset level, so that downstream controls can be applied automatically.
Classification schemes exist in nearly every organisation as a policy document and in very few as something the platform can act on. The difference is whether the label is attached to the data as metadata rather than described in a spreadsheet.
Once labels are attached and propagate through lineage, controls become automatic rather than discretionary: masking applied to anything labelled personal, export blocked for anything labelled restricted, retention policy selected by class, residency enforced by jurisdiction tag, and alerts raised when a labelled column appears in an unexpected destination.
The labelling itself is the effort, and doing it manually across a large estate does not finish. Automated classification — pattern matching for structured identifiers, statistical profiling, and increasingly model-based detection — gets a reasonable proportion of the way, and needs human review for the ambiguous cases and for anything where the consequence of a wrong label is high.
The property that must be designed in is propagation. A column correctly labelled in the source is worthless if the derived table built from it carries no label, and the derived tables are where most access actually happens. Lineage-driven label inheritance is what makes classification hold across a platform rather than at its edge.