Source: Role-based access control for Azure OpenAI
The built-in Azure RBAC roles for Azure OpenAI (OpenAI User, OpenAI Contributor, Cognitive Services Contributor, Usages Reader) and what each can and cannot do.
Contents
- What: official how-to (applies only to the Foundry (classic) portal)
- Publisher: Microsoft
- Date: ms.date 2026-09-10; accessed 2026-11-14
- Original: Role-based access control for Azure OpenAI
Summary
Maps the four built-in roles to concrete capabilities, which shows how to split who calls the model from who deploys, tunes, or holds keys.
Key claims
- Roles can be assigned to users, groups, service principals or managed identities, at resource, resource-group, subscription or management-group scope.
- Cognitive Services OpenAI User: can run inference with Microsoft Entra ID and use the playgrounds. Cannot see keys, deploy models, fine-tune or read stored completions.
- Cognitive Services OpenAI Contributor: adds deployments, fine-tuning, dataset upload and stored-completions access. Still no keys and no quota.
- Cognitive Services Contributor: can create resources, view and regenerate keys, create custom guardrails, and also deploy and fine-tune. It cannot make Entra ID inference calls or access quota.
- Cognitive Services Usages Reader: quota visibility, subscription scope only.
- Subscription-level Owner and Contributor are inherited and take priority over the Azure OpenAI roles applied at resource-group level. The page warns that using API keys (its example is Azure AI Search admin or query keys) bypasses Azure RBAC.