Skip to content
Need to Know

GUIDE, read 13 Nov 2026

A classification travels only as far as the last tool that reads it and passes it on

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.

The exposure mapAll guides

Four kinds of tool create the first classification

The governance question is who writes the label and who reads it afterward. Warehouse scanners write it inside the platform. Snowflake says its classification discovers sensitive data and makes it easy to apply tags and masking policies. Databricks says that with automatic tagging on, all existing and future detections get class tags, and a reviewed result can seed a masking policy. Immuta's identification runs inside the data platform and tags by column name, pattern or dictionary.

Data security posture tools, such as Cyera and BigID, scan from outside. Microsoft documents more than 200 built-in classifications in the Purview Data Map. Catalogs add a fourth source: Select Star matches PII rules against column names once a day, which is name-based, not a scan of values. Each source writes into a different place, and a downstream tool can only carry what it can read.

Tools differ on whether they take classifications in from elsewhere

Several ranked tools import labels rather than create them. Atlan says its Cyera connector imports Cyera classifications onto assets and that it syncs tags from Snowflake, Databricks, BigQuery and dbt. Secoda says its Cyera integration tags matching resources with the same classifications. Immuta says it ingests tags from Alation, Atlan, Collibra and Microsoft Purview, polling the catalog every 24 hours. Privacera's docs describe a Collibra connector that keeps tags synchronized. Informatica says its governance catalog can extract Unity Catalog tags, and its page does not say they are then propagated along lineage.

Euno documents Cyera and BigID imports and then carries the labels forward through column-level lineage. Per Euno's docs, BigID classifications do not appear on Looker, Tableau or Power BI assets on their own, and Euno propagates them to downstream fields and reports. Euno reads documented at the column stage and documented at the BI field stage.

Snowflake's own classification tags are the ones its propagation skips

This is the one check most Snowflake teams can run in ten minutes. Snowflake's classification page says it marks classified columns with two system-defined tags, SNOWFLAKE.CORE.SEMANTIC_CATEGORY and SNOWFLAKE.CORE.PRIVACY_CATEGORY. Snowflake's propagation page lists its limits, and the first one reads "System tags are not propagated" (both pages read 13 Nov 2026). Read together, the documents say the label a scan writes does not itself travel to a view or table built from the column.

The route the classification page documents is a tag map: when a system tag takes a value, a user-defined tag is applied as well. Propagation can then carry the user-defined tag. The same page gives a second limit: one transaction propagates a tag to at most 10,000 downstream objects. A team that turned on classification and propagation and never built the map has a gap that no report will show. Test it on one view before an audit asks.

Past the warehouse, every ranked tool carries it a different distance

Databricks says tag propagation (Beta) copies a governed tag from a source table or column to derived tables and views, and that a propagated tag makes a matching column mask apply to the derived column. Propagation is off by default and enabled per governed tag. Atlan likewise says propagation is off by default and chosen per tag. Microsoft documents label inheritance from data sources to Power BI semantic models for only three source types, and its downstream inheritance stops above 80 items.

Immuta's docs say its policies attach by tag to tables made by dbt, but that it receives no DDL events, which leaves a window it calls policy downtime. DataHub's term propagation covers datasets and excludes charts and dashboards. The table counts the ranked tools by the cell each reads on the exposure map.

StageDocumentedPartialNone found
Column (classification exists)14 of 162 of 160 of 16
Transformation model1 of 1612 of 163 of 16
BI field3 of 1610 of 163 of 16
Report4 of 168 of 164 of 16
Viewer0 of 1614 of 162 of 16
Agent6 of 1610 of 160 of 16

The chain usually breaks at the BI field and the agent's sentence

The counts show where the chain breaks. The column stage is the fullest, with 14 of 16 documented. The BI field and report stages are where Immuta, Privacera and OneTrust read none found, none found and none found: their controls sit on the data, and the label does not follow it into the dashboard. The agent stage is the second break. 6 of the 16 tools read documented there, for example Snowflake Horizon and Databricks Unity Catalog, because the agent queries under enforced policies, but a policy enforced on the query does not label the sentence the agent writes afterward.

The practical reading is that a classification is a claim made at one stage and re-made at each later one. Each re-making needs a named tool, a setting that is switched on, and a place to look when it fails. Where none of the three exists, treat the downstream field as unclassified, whatever the warehouse says.

Check three things in a trial. Ask the vendor to show a classified column and its label on the BI field. Ask what happens to a label when a model renames or joins the column. Ask whether the agent's tool response carries the label or only obeys the mask. See the exposure map for each tool's row, and the glossary entry on label inheritance.

Sources

  1. Snowflake classification, read 13 Nov 2026
  2. Snowflake tag propagation limits, read 13 Nov 2026
  3. Databricks tag propagation, read 13 Nov 2026
  4. Databricks Data Classification, read 13 Nov 2026
  5. Microsoft downstream label inheritance, read 13 Nov 2026
  6. Microsoft label inheritance from data sources, read 13 Nov 2026
  7. Immuta external catalog tags, read 13 Nov 2026
  8. Immuta dbt transform workflow, read 13 Nov 2026
  9. Euno docs, BigID use cases, read 13 Nov 2026
  10. DataHub term propagation, read 13 Nov 2026

Last reviewed 13 Nov 2026.