Databricks AI
Databricks · Model API · US · www.databricks.com
services/databricks-ai.yamlVendor data-handling terms
“For all paid accounts, Model Serving does not use user inputs submitted to the service or outputs from the service to train any models or improve any Databricks services.”
“Your inputs and outputs are isolated from those of other customers, stored in the same region as your workspace for up to thirty (30) days, and only accessible for detecting and responding to security or abuse concerns. / Partner model providers may retain data for safety purposes.”
The Databricks Geos article states cross-Geo processing is enabled by default for workspaces in Geos outside the US and EU without a compliance security profile.
“Foundation Model APIs is a Databricks Designated Service , which means that it uses Databricks Geos to manage data residency when processing customer content.”
Domains and endpoints
observed means seen in Unseen deployments; vendor-documented means listed by the vendor. Vendors do not publish complete lists.
| Host | Role | Source |
|---|---|---|
| api.databricks.com | api | observed |
Assessment
Reasoning: Model serving and foundation model APIs inside a customer's Databricks workspace. No training on inputs for paid accounts, 30-day abuse-review retention in the workspace region, and Geo-bound processing that admins can loosen. Partner-hosted models (OpenAI, Anthropic) add their own safety retention.
Flags: Geo-bound processing, admin-configurable · Partner model retention applies · Assessed 2026-09-16. The assessment is Unseen's; the terms above are the vendor's.
Changelog
- 2026-09-16Documented. Training, retention and residency terms verified against the Databricks Model Serving and Foundation Model APIs documentation.
- 2026-09-15Listed from the Unseen catalogue with observed domains.
Corrections
Pull request on GitHub, or the form below. Changes are reviewed and recorded in the changelog.