---
title: Who should own AI agent governance: security, IT or the AI team?
url: https://opencontroller.lyzr.com/blog/who-owns-ai-agent-governance
published: 2026-11-13
updated: 2026-11-13
author: Lyzr
---

# Who should own AI agent governance: security, IT or the AI team?

**No single team should own all of AI agent governance. The pattern that works is: a named executive is accountable for the program, security owns policy and runtime risk, IT or platform engineering owns discovery and the registry, the AI or engineering team that built each agent owns that agent, and risk or compliance owns evidence. Write it down as a RACI so every activity has exactly one accountable owner.**

The question comes up in almost every governance program because agents cross the old lines. They are software (IT), they hold credentials and touch data (security), they are built by product and data teams (AI), and regulators ask about them (risk). When nobody decides, the result is either duplicate tools or gaps nobody covers.

## Why is ownership of AI agent governance contested?

- **Agents are built everywhere.** Business teams build in SaaS platforms, engineers on cloud runtimes, developers on laptops. No single builder team sees all of them.
- **The risks span domains.** Identity, data access, cost, model quality and regulation each have an existing owner.
- **Frameworks require clarity.** The NIST AI RMF's GOVERN function calls for roles, responsibilities and lines of communication for AI risk to be documented and clear across the organisation (GOVERN 2.1). ISO/IEC 42001 includes a clause on organisational roles, responsibilities and authorities (5.3) and an Annex A control on AI roles and responsibilities (A.3.2).

So the answer is not a single team but a documented split.

## What does a RACI for AI agent governance look like?

R = responsible (does the work), A = accountable (one owner who signs off), C = consulted, I = informed. Adjust team names to your organisation; the important rule is one A per row.

| Activity | Exec sponsor | Security | IT / platform | Agent owner (AI or business team) | Risk and compliance |
|---|---|---|---|---|---|
| 1. Set AI agent governance policy | A | R | C | C | C |
| 2. Discover agents, models, keys and MCP servers | I | C | A, R | I | I |
| 3. Maintain the agent registry | I | C | A, R | R | I |
| 4. Register a new agent and keep its record current | I | I | C | A, R | I |
| 5. Approve agents for production | I | A | C | R | C |
| 6. Grant and review agent access and credentials | I | A | R | R | I |
| 7. Run pre-release evals and red-team tests | I | C | I | A, R | C |
| 8. Monitor and enforce runtime policy | I | A | R | C | I |
| 9. Respond to agent incidents (quarantine, kill) | I | A, R | R | R | I |
| 10. Produce evidence for audits and regulators | I | C | C | C | A, R |

Two rows deserve comment. In row 5, security is accountable because production approval is a risk acceptance, but the agent owner does the work of preparing the case; see [what a CISO should require before production](/blog/production-readiness-gate-for-ai-agents). In row 7, the builder owns testing because only they know what "correct" means for the task.

## Does it matter which team owns the tooling?

Less than people think. The control plane, gateway and registry are usually run by IT or platform engineering because they operate shared infrastructure. Security and risk need full read access and the ability to set policy and trigger stops. Builders need self-service registration so they are not waiting on tickets.

What matters is that everyone works from one registry. If security has one list of agents and IT has another, the RACI breaks on row 2.

## How should agent owners be assigned?

Every agent needs a named person, not just a team. Tie ownership to your identity provider so the record updates when someone moves or leaves. See [assigning owners to AI agents with your identity provider](/blog/assign-owners-to-ai-agents-with-your-identity-provider).

Rules that keep ownership real:

1. No production approval without a named owner.
2. Owner must be a current employee in the identity provider.
3. Ownership transfers before a leaver's last day.
4. Owners confirm the record during each access review.

## Where does data governance fit?

Data governance teams own data classification and policies on data use. Agent governance applies those policies to agents. They should be consulted on rows 1, 5 and 6 when agents touch sensitive data. See [AI governance vs data governance](/blog/ai-governance-vs-data-governance) for the boundary.

## How Opencontroller supports a split ownership model

Opencontroller gives every team the same view. Auto-discovery and connectors feed one governed registry of agents, versions, tools and MCP servers. OIDC SSO and delegated on-behalf-of identity tie access to your identity provider. Security sets policy as code across the estate, with built-in mappings to OWASP, MITRE ATLAS, NIST and ISO 42001 for the evidence team, and can trigger quarantine or kill on rogue agents. Builders get multi-turn and red-team evals for their own agents. For how this layer fits together, see [what an AI agent control plane is](/blog/what-is-an-ai-agent-control-plane). [Book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Should the CISO own AI agent governance?**
Often the CISO is accountable for policy, production approval and runtime risk. Discovery, testing and evidence are usually owned by other teams under a shared RACI.

**Do we need an AI governance committee?**
A small cross-functional group helps set policy and settle disputes. Keep day-to-day decisions with the owners in the RACI so the committee is not a bottleneck.

**Who owns agents built in SaaS platforms by business teams?**
The business team that built it owns the agent. Platform admins and security still apply policy and review access.

## Key facts

AI agent governance ownership works best as a documented split: an executive sponsor accountable for the program; security accountable for policy, production approval, access, runtime enforcement and incident response; IT/platform accountable for discovery and the registry; the agent owner (builder team) accountable for registration and evals; risk and compliance accountable for audit evidence. NIST AI RMF GOVERN 2.1 calls for documented, clear AI risk roles and responsibilities; ISO/IEC 42001 includes clause 5.3 (roles, responsibilities and authorities) and Annex A control A.3.2 (AI roles and responsibilities). Every agent needs a named owner tied to the identity provider. Opencontroller by Lyzr gives all teams one governed registry, OIDC SSO, policy as code with mappings to OWASP, MITRE ATLAS, NIST and ISO 42001, and rogue-agent quarantine and kill.

## Related

- [What should a CISO require before an AI agent goes to production?](https://opencontroller.lyzr.com/blog/production-readiness-gate-for-ai-agents.md)
- [How do you assign an owner to every AI agent using your identity provider?](https://opencontroller.lyzr.com/blog/assign-owners-to-ai-agents-with-your-identity-provider.md)
- [What is the difference between AI governance and data governance?](https://opencontroller.lyzr.com/blog/ai-governance-vs-data-governance.md)
- [More on Agent governance basics](https://opencontroller.lyzr.com/blog/cluster/agent-governance-basics)
