Hello @mark_o , I took a look at both internal and external documentation and here is what I found.
Short answer: yes. Databricks states that Data Classification supports all catalog types, and a foreign catalog is one of them. The statement is in the Data Classification release notes under the Public Preview entry dated October 13, 2025, and nothing in the GA docs walks it back. The current Limitations section only covers views.
The reason you couldn't find a clear statement is that the main docs page never says it directly, and the early Beta docs actually said the opposite ("standard catalogs only"). That restriction was lifted at Public Preview, so if you hit an older blog or cached page that says no, that's why.
A few things to keep in mind on a foreign catalog:
- A foreign catalog is a read-only mirror of an external database or catalog. Classification reads the data in place over Lakehouse Federation; nothing is migrated and the source system isn't changed. The results and any
class.* tags are Unity Catalog governance metadata on the Databricks side.
- The scan runs on serverless compute, so the external source has to be reachable from serverless, same as a federated query from a serverless SQL warehouse. If your connection only works from classic compute inside your network, sort that out first.
- Expect the scan to put query load on the source system, since that's where the data lives. The Schema scope option in the enable dialog lets you limit scanning to the schemas you care about, and the initial scan costs more than the incremental ones that follow.
- Permissions are the same as any catalog: own it or hold
MANAGE to enable; USE CATALOG plus APPLY TAG on the catalog and ASSIGN on the tag for auto-tagging.
To enable it, open the catalog in Catalog Explorer, go to the Details tab, and click Enable next to Data Classification, or use Configure on the Data Classification results page and pick the catalog from the list. Start with one schema and check the results in the UI or in system.data_classification.results before widening the scope. If the catalog doesn't show up as an option, or a scan won't start, that's a deployment-specific problem (connector type, cloud, region, permissions) and the right move is a support ticket or a note to your account team.
Plan B if you'd rather not scan the source system directly: build a materialized view over the foreign tables in a standard catalog and classify that. Materialized views are in scope by default, and the federation docs recommend this pattern for loading external data:
CREATE MATERIALIZED VIEW my_catalog.my_schema.customers_mv AS SELECT * FROM my_foreign_catalog.source_schema.customers;
References
If you enable it, please post back with what you see. Real-world results on a foreign catalog would help the next person who searches this.
Regards,
Louis
