Skip to main content

Buyer guide

Databricks Genie Governance & AI Data Checklist

Unity Catalog permissions, embedded warehouse credentials, model partners, data sent to models, retention, Geos, and audit logs for Genie One and Genie Agents.

Reading time
2 min read
Sources
10 cited

Review Genie's data handling before opening access. The SQL Genie generates is always read-only and runs under the asker's Unity Catalog permissions. Partner models receive prompts and metadata with no retention. Databricks encrypts stored responses in its control plane.

When you need this

  • Security review asks whose permissions a Genie answer runs under.
  • Legal asks which model providers see company data and whether anything trains on it.
  • You are about to give business users consumer access and Genie One.

Evaluation criteria

  1. Data access

    Whose permissions apply to an answer?

    Look forCompute runs on the author's embedded warehouse credentials, and data access is checked as each end user, with row filters and column masks enforced per user. A question about data the user cannot access returns an empty response.

  2. Who can build and schedule

    Which entitlements do business users get?

    Look forConsumer access is enough for Genie One chat. One team found that scheduling required Databricks SQL access, which also lets users create agents and dashboards, so decide who authors before granting it.

  3. Model partners

    Which models answer, and can you turn them off?

    Look forChat mode runs on Azure OpenAI and OpenAI on Databricks, and Agent mode on OpenAI and Anthropic on Databricks. Neither works when partner-powered AI features are off. Databricks had named November 1, 2026 to fully remove that setting and now says it will do so at a later date. Once it is removed, changing the setting means contacting your Databricks account team.

  4. What leaves the platform

    What data is sent to models, and is it kept?

    Look forGenie sends the question, table names and descriptions, relevant column values, instructions, and example SQL, filtered by Unity Catalog permissions. Partner providers retain none of it, and Databricks does not train foundation models on it.

  5. Residency

    Where is the work processed?

    Look forAI assistive features are Designated Services under Databricks Geos. Agent mode runs in-Geo in the Americas, Asia, Australia and New Zealand, Europe, and India, and needs cross-Geo processing only in Mainland China and the Middle East and Africa.

  6. Audit and review

    Who can see conversations, and what is logged?

    Look forAgent managers see users' questions, and the full exchange for conversations set to Reviewable. Audit log queries cover feedback, review requests, and API calls under the aibiGenie service.

Questions to ask vendors

  1. Are our Unity Catalog grants clean enough that Genie should reproduce them as they are?
  2. Is partner-powered AI enabled on every workspace that will use Genie, and is that our decision?
  3. Does our compliance profile allow Agent mode in our region?
  4. Who reviews flagged conversations, and who can delete them?

Red flags

  • Opening Genie One before cleaning up Unity Catalog grants.
  • Granting Databricks SQL access to everyone so they can schedule questions.
  • Assuming Genie never sees data values, when relevant column values are sent as context.
  • No owner for the audit queries on feedback and review requests.

Products in this piece

Sources

Official documentation
8
Public source
2
  1. Databricks docs: How Genie Agents workOfficial documentationdocs.databricks.com
  2. Databricks docs: Create and manage a Genie AgentOfficial documentationdocs.databricks.com
  3. Databricks docs: AI assistive features trust and safetyOfficial documentationdocs.databricks.com
  4. Databricks docs: Partner-powered AI featuresOfficial documentationdocs.databricks.com
  5. Databricks docs: Genie OneOfficial documentationdocs.databricks.com
  6. Databricks docs: Test and monitor a Genie AgentOfficial documentationdocs.databricks.com
  7. Databricks docs: Monitor Genie Agents with audit logsOfficial documentationdocs.databricks.com
  8. Databricks docs: Embed a Genie AgentOfficial documentationdocs.databricks.com
  9. Reddit post: scheduling Genie questions requires SQL authoring access (Jun 2, 2026)Public sourcereddit.com
  10. Mechanical Rock: Databricks Genie One opens the doors for business users (Aug 10, 2026)Public sourcemechanicalrock.io