GUIDE, read 13 Nov 2026
Databricks carries a classified field into derived tables and Genie, and a dashboard's default setting can run it as the publisher
Where Euno is behind
Euno is not a policy engine or a DSPM: it reads metadata only, so it does not scan data contents or mask at the warehouse, which leaves it behind Atlan on acting on findings (5 to 7), and it publishes no dollar prices.
Databricks links classification, tags and masks into one chain
Databricks says Data Classification tags columns automatically with class.* governed tags, covering existing and future detections, and that new tables and columns are typically classified within 24 hours. A classification result can be turned into a masking policy whose form is pre-filled with the reviewed tag. On 15 Sep 2026 Databricks added Beta scanning of Unity Catalog views. ABAC policies attach at catalog, schema or table level and support row filters and column masks, with DENY policies in Beta. In the exposure map, Databricks Unity Catalog reads documented at the column stage and documented at the agent stage.
The stage table shows what each Databricks layer covers
Each row is one stage between the warehouse column and the agent.
| Stage | What Databricks' own features do | Where it stops |
|---|---|---|
| Column | Data Classification writes class.* governed tags; ABAC masks and filters act on them | Classification of views is Beta; none found on 13 Nov 2026: a named integration that feeds third-party labels into Unity Catalog tags |
| Transformation | Tag propagation (Beta) copies a governed tag along lineage to derived tables and views, and a matching ABAC policy then applies | Serverless compute only; off by default and enabled tag by tag; custom ungoverned tags are not propagated |
| BI field | AI/BI dashboards appear as downstream consumers in lineage | Tableau and Power BI assets need manual registration as external metadata objects; none found on 13 Nov 2026: automatic harvesting |
| Report | Dashboard view events appear in the audit log | Scheduled-email PDF snapshots are excluded from the view event |
| Viewer | Audit log and a query history system table name the user | None found on 13 Nov 2026: viewer logs for Tableau or Power BI reports on Databricks tables |
| Agent | Genie enforces row filters and column masks per user; managed MCP calls are written to the audit log | The Genie Agent MCP bypasses Unity Gateway; Sensitive Data Detection on MCP is Beta |
The default dashboard publish setting changes who a mask is applied to
Databricks documents two publish settings for AI/BI dashboards. Under the default, Share data permissions, viewers run queries with the publisher's data permissions, and Databricks warns this might expose data to users who have not been granted direct access. Under Individual data permissions, each viewer runs queries with their own credentials, so a column mask or row filter evaluates for the person looking. The dashboard audit event carries a credentials_embedded parameter.
The practical reading: a mask that works in Genie, which enforces policies per user, can behave differently in a dashboard published with the default. For each dashboard that draws on a classified column, record which setting it uses. This is a check a governance lead can run from the dashboard's share settings and the audit log without buying anything. Databricks also says the audit event for opening a published dashboard covers workspace users and account users, so the log can show who looked even when the query ran as someone else. Whether the masked value was shown depends on the publish setting, not on the log.
The managed MCP audit event links a tool call to a SQL statement
Databricks' audit log documents a managedMcp service whose mcpRequest event records the server type (Databricks SQL, Unity Catalog functions, AI Search or Genie), the MCP method, the tool name, and, where relevant, a sql_statement_id and the Genie conversation and message IDs. Internal service-to-service requests are excluded. Genie One MCP calls are logged one event per tool call. Databricks says MCP clients authenticate as a user or service principal, and OAuth calls run as the signed-in user.
The useful join is mcpRequest to the query history system table through the statement ID, which ties an agent's request to the statement text and the user. Databricks announced on 26 Aug 2026 that statement text in system tables is masked by default, so an auditor needs the databricks_pii_access group or an account admin to read it.
Which ranked tools extend it
Databricks documents external lineage for Tableau and Power BI only by manual registration, so its BI stages read partial and partial. Tools that read the BI layer themselves close part of that gap. Alation documents column-level lineage for Tableau and Power BI (BI field documented, report documented). Select Star documents lineage and top users per dashboard (documented, documented). Atlan syncs Databricks tags and documents Tableau column-level lineage.
Informatica says it extracts governance tags from Unity Catalog and pushes policies into Databricks, Power BI and Tableau, and the page read does not say the tags are propagated along lineage. Immuta says its ABAC policies appear as native Unity Catalog controls and that it ingests Unity Catalog tags. Per Euno's docs, Euno, a context platform reading metadata and query logs only, maps Cyera classifications onto warehouse resources and traces them forward into BI fields and reports; its report cell reads documented.
Check three settings before an agent reads Databricks data
Confirm the tag propagation preview is on and the governed tag is enabled for propagation, since ABAC policies do not apply to a derived column until the tag appears there. Confirm each dashboard's publish setting. Confirm whether your agent reaches data through Unity Gateway or the Genie Agent MCP, because the second relies on permissions on the Genie Agent and its data instead.
Sources
- Databricks: Data Classification, read 13 Nov 2026
- Databricks: governed tag propagation, read 13 Nov 2026
- Databricks: attribute-based access control, read 13 Nov 2026
- Databricks: audit log reference, read 13 Nov 2026
- Databricks: share and publish AI/BI dashboards, read 13 Nov 2026
- Databricks: Genie set up, read 13 Nov 2026
- Databricks: managed MCP servers, read 13 Nov 2026
- Databricks: external lineage, read 13 Nov 2026
- Databricks release notes, August 2026, read 13 Nov 2026
- Databricks release notes, September 2026, read 13 Nov 2026
- Immuta docs: tag ingestion from catalogs, read 13 Nov 2026
Last reviewed 13 Nov 2026.