---
title: What is the difference between AI governance and data governance?
url: https://opencontroller.lyzr.com/blog/ai-governance-vs-data-governance
published: 2026-11-11
updated: 2026-11-11
author: Lyzr
---

# What is the difference between AI governance and data governance?

**Data governance controls the data itself: what it means, who owns it, how good it is and who may read it. AI governance controls the systems that act on data: which models and agents exist, what they are allowed to do, how they behave in production and what evidence shows they were controlled. The two overlap at one point — what data an agent can reach and send to a model — and that is where they must share identifiers.**

Buyers often ask for data dictionaries, glossaries and data asset registers in the same breath as agent governance. That is reasonable. An agent that reads a customer table is only as well governed as the table's classification. But a data catalogue cannot tell you that an agent exists, what tools it calls, or whether it has started behaving differently since Tuesday. This post draws the boundary.

## What does data governance cover?

Data governance is the older discipline. Its core objects are datasets, columns, glossary terms and data products. Typical controls are:

- **Definitions:** a business glossary and data dictionary, so "active customer" means one thing.
- **Ownership:** a data owner and a steward for each domain or dataset.
- **Classification:** labels such as public, internal, confidential or personal data.
- **Quality:** completeness, freshness and accuracy rules with thresholds.
- **Access:** role- or attribute-based policies on who can read which tables and columns.
- **Lineage:** where data came from and where it flows.

Its tools are data catalogues, quality monitors and the access-control layer of the warehouse or lakehouse.

## What does AI governance cover?

AI governance treats models, agents, prompts, tools and MCP servers as the governed objects. The [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) organises it into four functions — Govern, Map, Measure and Manage — and [ISO/IEC 42001](https://www.iso.org/standard/42001) defines it as a management system with roles, risk assessment and continual improvement. In practice, for agents, it means:

- **Inventory:** every agent, model, version, tool and API key, including ones nobody registered. See [how to find shadow AI agents](/blog/how-to-find-shadow-ai-agents).
- **Identity and ownership:** who the agent acts as, on whose behalf, and which human is accountable.
- **Permissions:** which tools and actions the agent may use, separate from which data it may read.
- **Runtime behaviour:** guardrails, routing, redaction, blocking, quarantine.
- **Quality and safety evaluation:** evals, red-teaming and scoring of production traffic.
- **Evidence:** records that map controls to frameworks such as OWASP, MITRE ATLAS, NIST and ISO 42001.

The governed object changes, so the tool changes. That tool is usually an [AI agent control plane](/blog/what-is-an-ai-agent-control-plane).

## Which controls belong where?

| Control | Data governance (catalogue / warehouse) | AI governance (agent control plane) | Shared identifier |
|---|---|---|---|
| Business glossary, data dictionary | Owns it | Reads it for context | Glossary term ID |
| Data classification labels | Owns it | Enforces it on prompts and outputs | Classification label |
| Table and column access | Owns it | Checks the agent's identity is mapped | Service principal / workload identity |
| Agent and model inventory | — | Owns it | Agent ID, version |
| Tool and MCP server permissions | — | Owns it | Tool / server ID |
| PII redaction in prompts | Supplies the definitions | Enforces in the traffic path | Classification label |
| Lineage of data into a model call | Owns dataset lineage | Owns call-level traces | Dataset ID in trace attributes |
| Production behaviour monitoring | — | Owns it | Agent ID |
| Ownership | Data owner per dataset | Agent owner per agent | Employee ID from the IdP |
| Audit evidence | Data access logs | Agent action logs, eval results, policy decisions | Request / trace ID |

The right-hand column matters most. If the catalogue calls a dataset one thing and the agent trace calls it another, nobody can answer "which agents touched confidential data last month?"

## Where do the two have to meet?

Three join points come up in almost every enterprise.

1. **Classification at the prompt.** A column marked as personal data in the catalogue should trigger redaction or blocking when its values appear in a prompt. The label is defined once, in data governance, and enforced in the AI traffic path. See [how to redact PII at the gateway](/blog/how-to-redact-pii-in-llm-prompts-at-the-gateway).
2. **Identity at the data layer.** An agent that queries a warehouse uses some principal. If it uses a shared service account, data access logs cannot say which agent read what. Give each agent its own workload identity, or pass the user's identity through on-behalf-of delegation.
3. **Ownership from one source.** Both data owners and agent owners should resolve to the same employee record in your identity provider, so leavers are handled once.

## Do you need two separate programmes?

Usually you need one governance council and two control layers. The council sets policy: risk appetite, classification scheme, approval rules. The data layer and the AI layer each enforce it on their own objects. Organisations that try to stretch a data catalogue to cover agents end up with a static list of agents and no runtime control. Organisations that ignore the catalogue end up with an AI layer that cannot tell sensitive data from public data.

A practical checklist for the boundary:

1. Agree one classification scheme and publish it from the catalogue.
2. Require every agent to have an owner who exists in the IdP.
3. Give every agent a distinct identity at the data layer.
4. Put dataset IDs and classification labels into agent trace attributes.
5. Report on the join: agents × datasets × classification, monthly.

## How Opencontroller does this

Opencontroller is the AI layer, not a data catalogue. It auto-discovers agents, models and API keys across AWS, Azure and Google Cloud, keeps a governed registry of agents, versions, tools and MCP servers, and supports delegated on-behalf-of identity and agent workload identity federation. Policy is written as code with built-in mappings to OWASP, MITRE ATLAS, NIST and ISO 42001, so policy decisions become evidence. It uses open standards (OIDC, SAML, SCIM, OpenTelemetry, REST and webhooks) so identifiers can be shared with the systems that already govern your data. To see it on your estate, [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Is AI governance part of data governance?**
They overlap but are not the same. Data governance governs datasets; AI governance governs models and agents and how they behave. Most organisations run them under one council with two control layers.

**Can a data catalogue act as an AI agent inventory?**
Only as a static list. A catalogue does not discover agents, see their tool calls or enforce anything at runtime, which an agent control plane does.

**What should data and AI governance share?**
A classification scheme, employee IDs for owners, and identifiers for datasets, agents and requests, so you can report which agents touched which classes of data.

## Key facts

Data governance governs datasets (definitions, ownership, classification, quality, access, lineage). AI governance governs models and agents (inventory, identity, permissions, runtime behaviour, evaluation, evidence). They join on classification labels, agent identities at the data layer and owner employee IDs. Opencontroller is an AI agent control plane by Lyzr, not a data catalogue. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [What is an AI agent control plane?](https://opencontroller.lyzr.com/blog/what-is-an-ai-agent-control-plane.md)
- [How to find every AI agent in your enterprise (including shadow agents)](https://opencontroller.lyzr.com/blog/how-to-find-shadow-ai-agents.md)
- [How do you replace AI governance spreadsheets and questionnaires with continuous evidence?](https://opencontroller.lyzr.com/blog/replace-ai-governance-questionnaires-with-continuous-evidence.md)
- [More on Agent governance basics](https://opencontroller.lyzr.com/blog/cluster/agent-governance-basics)
