
When each document in an Agent Search data store is restricted to specific people, the search service must filter results by the signed-in user's permissions before returning them. Asking the model to drop restricted results leaves enforcement to the model rather than to the search service. On the Professional Agentic Architect exam, Agent Search document-level access control focuses on how data source access control works with Cloud Storage metadata, how third-party identities are recognized, and why other access controls solve different problems.
Data source access control lets Agent Search identify the signed-in end user through the identity provider and return only the documents that user is allowed to read. Because the search service performs this filtering directly, unauthorized documents are excluded from the results returned for that user.
For Cloud Storage data, each document carries ACL metadata listing its reader principals. Agent Search uses those reader principals alongside the signed-in user's identity to determine which documents the user may read.
Data source access control must be turned on when the data store is created, because it cannot be enabled later. Planning per-document permissions before creating the data store is required whenever document-level filtering is needed.
End-user identities do not always reside in Google. When user identities live in a third-party identity provider such as Okta and are not synced to Google, Agent Search needs a way to recognize those identities before it can apply document permissions.
Workforce Identity Federation enables Agent Search to recognize unsynced third-party identities. Pairing Workforce Identity Federation with a data store created with data source access control and Cloud Storage ACL metadata allows Agent Search to identify the signed-in user and return only the individual documents that user is permitted to read.
For the Professional Agentic Architect exam, it helps to separate document-level access control in Agent Search from controls that operate at a different layer or granularity.
The common thread is that document-level permissions are enforced by the search service at retrieval time, not by the agent or the model. That makes two decisions worth settling early: how Agent Search will recognize each user, and whether the data store needs data source access control, since that setting cannot be added after the data store exists.
Our Professional Agentic Architect practice exams test Agent Search document-level access control alongside Workforce Identity Federation and Cloud Storage ACL metadata, with detailed explanations that drill these distinctions.