authorization

Authorization for External Data Sources

Security Requirement

It MUST be verified that the principal accessing an external data source has permissions to access or modify the data. Authentication alone (knowing who the principal is) is not sufficient; authorization (what they can do) must also be enforced.

Authorization Scope

Operation	Authorization Check
Read data	Does the principal have read permission for this specific data?
Write data	Does the principal have write permission?
Modify schema	Does the principal have administrative permissions?
Delete data	Does the principal have delete permission?
Bulk export	Does the principal have permission for bulk data access?

Principle of Least Privilege

Each AI system component should have only the minimum data access permissions needed:

Component	Typical Permissions
RAG ingestion pipeline	Read-only on source data
Inference engine	No direct data source access (goes through RAG)
Agent with data tools	Read-only unless write is explicitly required
Admin/maintenance jobs	Broader access, time-limited and audited

Implementation Checklist

• Define the minimum data access permissions needed by each AI component
• Configure data source access controls to enforce least privilege
• Separate read and write permissions (do not grant write when only read is needed)
• Verify authorization on every data access request (not just at connection time)
• Regularly audit data source permissions and remove unnecessary access
• Use role-based or attribute-based access control at the data source level