5.3.1. Data Classification Capabilities
💡 First Principle: You can't protect sensitive data you don't know exists, so Purview's data classification capabilities automatically scan and identify content containing sensitive information — using built-in sensitive information types (like credit card number patterns) and trainable classifiers (machine-learning models trained to recognize categories like resumes or contracts) — turning "we think we might have sensitive data somewhere" into an actual inventory.
Sensitive information types use pattern matching (regular expressions, checksums) to detect structured data like national ID numbers or credit card numbers, while trainable classifiers handle less structured categories that a simple pattern can't reliably capture, learning from sample documents provided during training.
⚠️ Exam Trap: Classification identifies and labels sensitive content — it doesn't, by itself, restrict access or apply protection. That's what sensitivity labels (5.3.3) and DLP (5.3.4) do, often working from classification's findings.
Reflection Question: Why would an organization need a machine-learning-based trainable classifier for something like "resumes," rather than a simple pattern-matching rule the way credit card numbers are detected?
