CONCEPT Cited by 1 source
Agent governance via identity¶
The principle that AI agents should be governed by the same identity and access-control boundary as the humans they act on behalf of, rather than holding separate bulk credentials or bypassing the catalog with shared backend tokens.
Mechanism¶
Identity flows end-to-end:
- User authenticates via enterprise IdP (e.g. Azure AD)
- Agent receives a scoped token via OAuth 2.0 token exchange (on-behalf-of)
- Agent calls tools/data through Data Hub or MCP server
- Unity Catalog (or equivalent governance layer) makes the final authorization decision against the user's identity
- If the user lacks access to a table, masked column, or governed view, the agent receives the same boundary
"An agent can reason over the same product context a user sees, and it can call tools on the user's behalf, but it cannot become a production writer or bypass Unity Catalog with a shared backend credential." (Source: sources/2026-07-21-databricks-why-rd-data-belongs-in-the-lakehouse-and-why-agents-need-it)
Structural property¶
The operating model separates production ownership (engineers change pipelines/product definitions through reviewed code and governed deployment paths) from consumption (agents and users query data through governed surfaces). Agents are consumers, not producers.
Seen in¶
- sources/2026-07-21-databricks-why-rd-data-belongs-in-the-lakehouse-and-why-agents-need-it — cellcentric's Data Hub on Azure + Databricks; Azure AD → OAuth 2.0 token exchange → Unity Catalog final authorization.
- sources/2026-05-20-databricks-governing-ai-agents-at-scale-with-unity-catalog — Databricks' "four pillars of agent governance" framing.
Related¶
- patterns/on-behalf-of-agent-authorization — the operational pattern
- concepts/four-pillars-of-agent-governance — the broader governance framework
- concepts/governed-agent-data-access — agents within governance boundary