Quick answer
Identity propagation is an optional query-time governance control, not a prerequisite for the Amazon Quick Agentic Catalog Experience. It forwards the signed-in user’s identity to the upstream authorization system so the source can enforce that user’s permissions.
It works only for DirectQuery Datasets. If an author switches a generated Dataset to SPICE or adds transformations, identity propagation does not apply.
| Catalog | Propagation path | Upstream enforcement |
|---|---|---|
| AWS Glue Data Catalog | Trusted identity propagation through AWS IAM Identity Center | Lake Formation, with the associated Athena and S3 access path |
| Databricks Unity Catalog | OAuth 3LO | Databricks permissions for the end-user identity |
Without propagation, use and maintain Quick row-level security (RLS) and column-level security (CLS) where appropriate.
Keep four controls separate
- Catalog connection authentication lets Quick and the author discover metadata.
- Author permissions control which assets can be curated and shared.
- Identity propagation controls source enforcement for each viewer at query time.
- Quick RLS or CLS applies Quick-managed row or column restrictions when designed.
A working service-role, PAT, or OAuth connection proves only that the connection works. It does not prove every viewer receives the right rows and columns.
Glue trusted identity propagation
AWS documents these major prerequisites:
- the Quick account uses AWS IAM Identity Center;
- Glue databases and tables are governed by Lake Formation;
- Lake Formation is integrated with the same Identity Center instance;
- users or groups have the correct Lake Formation grants and filters;
- S3 data locations are registered as required;
- the Athena results bucket has the necessary S3 Access Grants;
- the Quick IAM role includes the required permissions and
sts:SetContexttrust.
Full-table permissions can override an intended filter. Test effective grants, not just the presence of row, column, or cell filters.
Databricks OAuth 3LO
The catalog integration supports PAT or OAuth 3LO for the agentic experience, but OAuth 3LO is the documented path for end-user identity enforcement. Test token consent, refresh, revocation, group changes, workspace and catalog permissions, and failure behavior.
Do not fall back from an expired user token to a broader shared identity without an explicit, reviewed policy.
For the catalog decision, see Glue versus Databricks.
DirectQuery is part of the security contract
Generated Datasets use DirectQuery by default. That makes upstream enforcement possible at query time. Editing the Dataset to add transformations or changing it to SPICE is not just a performance choice: AWS states that identity propagation then stops applying.
Create a deployment guard that detects this change and requires a fresh governance review. If the customized Dataset still needs user-level restrictions, design and test Quick RLS and CLS before sharing it.
Permission test matrix
Use at least four identities:
- an authorized user with full domain access;
- a user restricted by row or geography;
- a user restricted by sensitive columns;
- a user with no access.
For each identity, test a dashboard, direct Dataset use, Topic Q&A, exports, drill-downs, cached views, and failure messages. Repeat after group changes, revoked access, semantic sync, Dataset edits, and Region or account changes.
Also verify audit logs can connect the Quick user, upstream identity, query, decision, and result without leaking credentials or restricted data.
Frequently asked questions
Is identity propagation required for the Amazon Quick agentic catalog experience?
No. AWS documents it as optional and additive. Discovery, bulk creation, and semantic inheritance work with any supported authentication type.
When does identity propagation apply?
It applies only to DirectQuery Datasets. Switching a Dataset to SPICE or adding transformations means identity propagation no longer applies.
What can I use without identity propagation?
AWS documents Quick RLS and CLS as manual alternatives. They require separate design, testing, and maintenance.
Official sources
- Amazon Quick: Identity propagation for data governance
- Amazon Quick: Glue trusted identity propagation
- Amazon Quick: Supported catalogs
Source check: August 4, 2026. Verify preview status, identity prerequisites, DirectQuery behavior, and effective permissions before sharing data.