---
title: "Governing AI agents at scale with Unity Catalog"
description: "With AI agents proliferating across organizational functions, governance must answer which agents access sensitive data and how they act. Traditional approaches either scatter logs and permissions across systems or restrict deployment, while agents require controls over autonomous tool use and runtime behavior. Unity Catalog and Unity AI Gateway address this through four pillars: delegated access, data-centric governance, cost intelligence, and open interoperability. The design passes user identity through agents with on-behalf-of tokens, evaluates MCP tool calls with Service Policies, applies inline guardrails, and records model, access, trace, and usage data in lakehouse tables. It also connects data quality, classification, spend, and outcomes, while supporting multiple frameworks and model providers through shared governance."
---

# Governing AI agents at scale with Unity Catalog

[Databricks](https://yomu.fyi/company/databricks) · David Nasi, Stefania Leone · May 20, 2026

**Type:** Explainer

## Summary

With AI agents proliferating across organizational functions, governance must answer which agents access sensitive data and how they act. Traditional approaches either scatter logs and permissions across systems or restrict deployment, while agents require controls over autonomous tool use and runtime behavior. Unity Catalog and Unity AI Gateway address this through four pillars: delegated access, data-centric governance, cost intelligence, and open interoperability. The design passes user identity through agents with on-behalf-of tokens, evaluates MCP tool calls with Service Policies, applies inline guardrails, and records model, access, trace, and usage data in lakehouse tables. It also connects data quality, classification, spend, and outcomes, while supporting multiple frameworks and model providers through shared governance.

## Context

Organizations with thousands of AI agents lack a single way to determine which agents access customer PII, correlate activity across systems, or attribute AI spending. Traditional governance was designed for predictable applications, while agents make autonomous choices and chain together tools. Overly permissive deployments create unmeasured risk, whereas extensive review can slow adoption and frustrate developers.

## Approach / What changed

Unity Catalog extends its permissions, lineage, and audit infrastructure to LLMs, MCP servers, skills, and agents, while Unity AI Gateway evaluates and logs model calls, tool invocations, and agent interactions. The approach combines delegated user access, MCP Service Policies, inline model guardrails, lakehouse-based audit and usage tables, data quality and classification, budgets, and open integrations across frameworks and model providers.

## Takeaways

- On-behalf-of token passing lets agents inherit the invoking user's Unity Catalog permissions instead of using shared service accounts; actions are logged against both the user and agent identities.
- Unity AI Gateway stores model-call payloads, token counts, latency, access operations, and usage data in lakehouse tables, enabling analysis of PII access, inference spend, data quality, and business outcomes.
- Governance is designed to follow governed data, services, and endpoints across frameworks and providers: MCP servers register once, while one policy and audit trail can cover multiple model destinations.

**Tags:** [AI Governance](https://yomu.fyi/topic/ai-governance), [MCP](https://yomu.fyi/topic/mcp), [Unity AI Gateway](https://yomu.fyi/topic/unity-ai-gateway), [Unity Catalog](https://yomu.fyi/topic/unity-catalog)

- Source: [Databricks](https://www.databricks.com/blog/governing-ai-agents-scale-unity-catalog)
- Source URL: https://www.databricks.com/blog/governing-ai-agents-scale-unity-catalog
- Ingested by Yomu: 2026-08-31T03:33:44.792Z

[Read original post](https://www.databricks.com/blog/governing-ai-agents-scale-unity-catalog)