---
title: How do you control which ChatGPT Enterprise users get agent features?
url: https://opencontroller.lyzr.com/blog/chatgpt-enterprise-rbac-for-agent-features
published: 2026-11-16
updated: 2026-11-16
author: Lyzr
---

# How do you control which ChatGPT Enterprise users get agent features?

**Use role-based access control (RBAC). Workspace Owners create custom roles with granular permissions, assign them to groups (ideally synced from your identity provider with SCIM), and set a restrictive default role for everyone else. Permissions are additive, so a user in several groups gets the union of their roles. Design roles so the default is narrow and each group adds only what it needs.**

Admins ask this when they want to pilot agent features, apps and connectors, or Codex with one team before the whole company. RBAC is the mechanism. The design choices are where rollouts go wrong.

## What does ChatGPT Enterprise RBAC give you?

- **Built-in roles.** Enterprise workspaces have Owner, Admin, Member and Analytics Viewer roles.
- **Custom roles.** Workspace Owners create custom roles with granular permissions. Admins can view and, where supported, update roles, but role creation and assignment sit with Owners.
- **Groups.** Groups can be created in ChatGPT or synced from your directory through SCIM. Roles are assigned to groups.
- **A default role.** A workspace-wide default applies to users with no custom role.
- **Additive inheritance.** A user in multiple groups receives the union of all their roles' permissions.
- **Plan scope.** OpenAI states that custom RBAC and available permissions depend on plan and workspace. ChatGPT Business does not include custom member RBAC or SCIM group synchronisation.

## Which agent-related features can you gate?

The exact list depends on your plan and workspace, so confirm in the Admin Console. Capabilities that OpenAI documents as role-controllable include:

| Capability | Typical control | Why it matters |
|---|---|---|
| Agent features | Allow or deny by role | Agents browse, act and run tasks with less supervision |
| Apps and connectors | Per app; for supported apps, all actions, read-only actions, or a custom set | Write actions change data in other systems |
| Codex | Access and internet access | Code execution and network reach |
| Workspace agents | Who can build, publish, share, schedule or configure | Shared agents run for many users |
| GPTs | Create, edit, share | Shared GPTs can carry instructions and connected actions |
| Projects and memory | Creation, sharing, memory | Data persistence across conversations |

OpenAI also documents group-specific credit limits, so a pilot group can have a separate monthly default.

## How should you design roles for a phased rollout?

1. **Make the default role narrow.** Because permissions are additive, a broad default cannot be taken away by a narrower group role. Start with chat and approved read-only apps.
2. **Sync groups from your identity provider.** SCIM groups update as people join and leave. With SCIM-managed users, remove access at the identity provider, otherwise a later sync can provision them again.
3. **Create one role per capability tier**, for example "Agent pilot", "Apps write", "Codex". Combine them through group membership rather than building one role per team.
4. **Use read-only app actions first.** Turn on write actions per app once you have logging and an owner for each connector.
5. **Restrict publishing separately from use.** Many users can use a shared agent while few can publish one.
6. **Review the union.** For a sample of users in several groups, check the effective permissions. Additive roles produce surprises.
7. **Log and review.** Use the Compliance API and audit logs to see who used which features; see [ChatGPT Enterprise Compliance API audit](/blog/chatgpt-enterprise-compliance-api-audit).

## What does RBAC not cover?

- **What agents do inside connected systems.** RBAC decides who can turn a connector on; the target system's permissions decide what it can touch.
- **Other platforms.** RBAC governs ChatGPT only, not Claude, Copilot, or agents built in your cloud.
- **Custom MCP connectors in depth.** See [controlling ChatGPT Enterprise connectors and MCP](/blog/control-chatgpt-enterprise-connectors-and-mcp).
- **Shared GPT and agent content.** See [governing ChatGPT Enterprise GPTs and agents](/blog/govern-chatgpt-enterprise-gpts-and-agents).

## How Opencontroller does this

Opencontroller connects to OpenAI Enterprise and Claude Enterprise workspaces, alongside agents across AWS, Azure, Google Cloud and platforms such as Salesforce, SAP, Databricks and HubSpot, and records them in one governed registry. Posture flags risky permissions and configuration, and policy as code, with mappings to OWASP, MITRE ATLAS, NIST and ISO 42001, lets you express one rule across platforms. Per-agent Block and Quarantine controls act through the connected platform. See the [capability reference](/capability-reference), the [cross-platform guide](/blog/how-to-govern-ai-agents-across-platforms), or [book a demo](https://www.lyzr.ai/book-demo/). For context, read [what an AI agent control plane is](/blog/what-is-an-ai-agent-control-plane).

## Frequently asked questions

**If a user is in two groups with different ChatGPT roles, which applies?**
Both. Role inheritance is additive, so the user receives the union of permissions from all their roles.

**Does ChatGPT Business support custom roles and SCIM group sync?**
No. OpenAI states that ChatGPT Business does not include custom member RBAC or SCIM group synchronisation.

**Can admins allow only read-only actions for an app?**
For apps that support action control, admins can allow all actions, read-only actions, or a custom set.

## Key facts

ChatGPT Enterprise built-in roles are Owner, Admin, Member and Analytics Viewer. Workspace Owners create custom roles with granular permissions and assign them to groups, which can be synced via SCIM; a workspace default role applies to users without a custom role. Role inheritance is additive (union of permissions). Custom RBAC and available permissions depend on plan and workspace; ChatGPT Business lacks custom member RBAC and SCIM group sync. For apps that support action control, admins can allow all, read-only or a custom set of actions. Workspace agent controls govern who can build, publish, share, schedule or configure agents. Opencontroller connects to OpenAI Enterprise and Claude Enterprise workspaces. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [How do you govern GPTs and workspace agents in ChatGPT Enterprise?](https://opencontroller.lyzr.com/blog/govern-chatgpt-enterprise-gpts-and-agents.md)
- [How do admins control connectors and custom MCP servers in ChatGPT Enterprise?](https://opencontroller.lyzr.com/blog/control-chatgpt-enterprise-connectors-and-mcp.md)
- [What does the ChatGPT Enterprise Compliance API let you audit?](https://opencontroller.lyzr.com/blog/chatgpt-enterprise-compliance-api-audit.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
