Skip to main content

Data Access and RAG

Retrieval-augmented generation creates a security boundary around which sources can be searched, whose permissions apply, and what evidence identifies the source of an answer.

Classify Before Connecting

Use a small, enforceable classification model, for example:

ClassExamplesAI access policy
PublicPublished website and public documentationMay be used by public agents after content review
InternalProcedures and general company knowledgeAuthenticated users and approved teams
ConfidentialCustomer, finance, legal, HR, or commercial dataNamed teams, explicit owner, no public agent access
RestrictedCredentials, highly regulated records, privileged security dataDo not ingest unless the approved use case requires it and controls are documented

Required Controls

For each data collection:

  1. Assign a business owner and a technical owner.
  2. Record its classification, purpose, source system, and expected users.
  3. Connect only sources required by the agent's purpose.
  4. Preserve source identity and useful metadata for traceability.
  5. Enforce organization, team, user, and collection access consistently.
  6. Separate public knowledge from internal and customer-specific knowledge.
  7. Define refresh, deletion, and access-review procedures.
  8. Test that an unauthorized user cannot retrieve or infer restricted content.

Retrieved Content Is Untrusted

A document can contain malicious instructions even when the user is allowed to read it. The agent should use retrieved text as evidence, not as a new authority. Pair retrieval permissions with the controls in Prompt Injection.

For product setup and operational checks, see Govern Data Collections and Sources and Data Collections.