44.1 Data classification and what can enter prompts

Overview and links for this section of the guide.

Data Levels

  • Public: Docs, marketing site. (Safe for all models).
  • Internal: Code, Slack messages. (Safe for Enterprise models with Zero-Retention agreements).
  • Confidential: PII, financial data. (Safe ONLY if redacted or if you have a BAA).
  • Toxic: Passwords, private keys. (NEVER send to a model).

The Policy

You need a "Data Egress Policy" that checks every API call. `if (contains_ssn(prompt)) block()`

Where to go next