Departments can view different documents.
The model answers only from retrieved evidence.
Choose two designs.
Apply user permissions before retrieval and test authorization across search, cache, citations, logs, and output to prevent re-exposure.
Detailed explanation
Similarity does not override access.
Later stages cannot re-expose data.
Only authorized data reaches the model.
It may already reach the model or logs.
Check official documentation
Compare the following points with the official documentation.
Review data access control and leakage prevention in Task 5.1 and metadata filters in the Bedrock Knowledge Bases query guide. For two users with different departmental access, outline checks for retrieval candidates, caches, citations, logs, and output.
Official references
What to confirm
Explain why retrieval conditions must come from trusted caller permissions and why later stages must be checked for re-exposure.
Key points
- ACL
- Pre-filter
- Cache
Notes
- Environment: Reference review of the official AIF-C01 guide and AWS documentation; no AWS resources or inference are required.
- A metadata filter alone is not a complete authorization system. Derive conditions from trusted permissions for the authenticated caller, and prevent callers from removing them.
- Review the correctness and freshness of access metadata and permission-aware cache isolation. No real user data is retrieved and no AWS operation is performed.
Foundation review
Access control list
A list of principals and the access they have to a resource.
Pre-retrieval filter
Applying authorization conditions before documents become search candidates.