Labeling information by sensitivity (public, internal, confidential, restricted) so AI tools, prompts, and retention policies match actual risk.
Data classification is the hinge between productivity and policy: it determines which tools are approved, what may enter an LLM, and what must stay on-prem or air-gapped. Weak labels create shadow AI workarounds when approved tools cannot handle the real workload.
Operational habit
If classification is unclear, default conservative until legal/security agrees—especially for code, credentials, and customer payloads.