Each row ends in the exposure trail: six joined dots for the warehouse column, the transformation, the BI field, the report, the viewer and the agent. Filled is documented, half is partial, open is none found on 13 Nov 2026. The full map carries every cell's evidence.
Totals are the weighted sum of eight scores from 0 to 10, divided by 10. A vendor's own claim scores at most 7 until a second source confirms it. Ties on total go to the higher score on the heaviest criterion, then to the name.
The top four, as scored
Rank 1
67.0
3 of 6 stages documentedagent told 7/10
Vendor-stated: Euno takes classifications from Cyera and BigID, carries them along column-level lineage into BI fields and reports, and scopes MCP context by persona. Last documented stage: report.
Read the profileRank 2
65.5
2 of 6 stages documentedagent told 7/10
Vendor-stated: Tag-driven purposes mask or deny sensitive columns, tags can propagate through lineage, and MCP runs under the asking user's persona with every call logged. Last documented stage: agent.
Read the profileRank 3
61.0
1 of 6 stages documentedagent told 7/10
Vendor-stated: Column-level lineage tracking, glossary term propagation to downstream lineage columns, and per-user OAuth logins for MCP. Last documented stage: agent.
Read the profileRank 4
55.5
2 of 6 stages documentedagent told 6/10
Vendor-stated: Masking, row access and tag-based masking are enforced at query time for people and agents alike, with ACCESS_HISTORY recording who touched which column. Last documented stage: agent.
Read the profile
Where each is strongest
- Euno: Carries classification downstream, 8 of 10. Imports Cyera and BigID classifications and propagates them along column lineage into BI; Cyera lists the integration independently.
- Atlan: Follows sensitive data into BI, 7 of 10. Column-level lineage documented for Tableau, Looker and Power BI with stated gaps (Tableau flows and embedded, Power BI partial SQL sources).
- DataHub: Follows sensitive data into BI, 8 of 10. Column lineage from Snowflake through dbt into Looker, Power BI and Tableau is documented, and dbt Labs independently describes column lineage to downstream dashboards; Tableau needs a high role.
- Snowflake Horizon: Acts on it, 7 of 10. Dynamic masking, row access and tag-based masking policies are documented as enforced at query time (some Enterprise Edition).
- Databricks Unity Catalog: Controls what an agent is told, 7 of 10. Row filters and masks enforce per user in Genie, OAuth calls run as the signed-in user and a PII guardrail is in Beta; one agent MCP bypasses the gateway.
- Microsoft Purview: Acts on it, 7 of 10. Protection policies, DLP and endpoint blocking enforce on labeled items, but not on Power BI reports or dashboards, and have scale limits.
- Alation: Follows sensitive data into BI, 7 of 10. Column-level lineage from warehouse columns to Tableau report fields, Power BI and Looker, plus dbt extraction, is documented; vendor docs only, Power BI with stated limitations.
- Select Star: Follows sensitive data into BI, 7 of 10. Column-level lineage from warehouse through SQL parsing into Looker, Tableau, Sigma and Power BI is vendor-documented; press quotes only the CEO on lineage, so capped at 7.
- Secoda: Follows sensitive data into BI, 6 of 10. Column lineage and impact analysis from warehouse column to Looker and Tableau dashboards, and dbt manifest support; only those two BI tools.
- Collibra: Acts on it, 7 of 10. Masking and row filters are applied from classifications and enforced in target systems, with data quality workflows handing off to Jira and ServiceNow.
- Immuta: Acts on it, 8 of 10. Tag-based column and cell masking, review flows and dashboard subscription policies are documented, and Databricks' partner page confirms row security and column masking as native controls.
- Ataccama ONE: Evidence beyond the vendor, 8 of 10. OpenLineage ecosystem listing and a PeerSpot reviews page with reviewer pricing summary are independent; analyst claims are vendor-stated and the listing is a logo only.
- Informatica: Evidence beyond the vendor, 9 of 10. Databricks partner and customer pages, KPMG and Point72 quotes and Microsoft documentation confirm it independently.
- Privacera: Acts on it, 8 of 10. Tag-driven masking and deny policies are documented on several platforms and Databricks' partner directory confirms row, column and tag policy control; the alerts page is placeholder text.
- Securiti: Acts on it, 6 of 10. Dynamic masking by policy, auto-remediation, assigned findings and agent rollbacks are stated by the vendor only.
- OneTrust: Acts on it, 6 of 10. Masking and row filtering pushed into Snowflake and Databricks, plus AI Guard redact and block; AI Guard is not recommended for high volumes.
Picks by situation
AI data governance for analytics means knowing where a sensitive field surfaces and what an agent was told
A governance lead is asked two things before an agent connects to analytics data: which reports and dashboards show a regulated field, and what the agent will be told when a given person asks it a question. Warehouse policies answer part of the first. Business intelligence tools hold the rest, in fields, reports and viewer lists that the warehouse never sees.
The 16 tools here come from five groups: warehouse-native governance, catalogs with governance features, access control platforms, privacy and security suites, and Euno, a context platform that reads metadata and query logs. They are scored on one question: how far does each follow the field, and what does it control on each agent call.
Guides, sectors and glossary