Do you give least-privilege access to your AI resources?

Most AI apps start with an API key, because it is the quickest way to get a first answer from a model. But a key gives full access to the resource and can't be limited to one user or one action. Nobody can tell from the logs who used it.

That's why every app and every person should get only the role they need on your AI resources, with API keys turned off.

Give each identity the smallest role

Figure: The built-in roles to reach for first

• Assign it on one resource - Scope the role to the AI resource or project, not the resource group or subscription
• Watch inherited roles - Owner and Contributor on the subscription are inherited and take priority over the narrow role
• Do the same for search - An app that only queries Azure AI Search gets Search Index Data Reader

❌ Figure: Bad example - The app holds a key with full access, and every developer can deploy models and change the whole resource group

✅ Figure: Good example - Each identity can do its own job and nothing more

Make the roles count by turning off keys

Roles only apply when you sign in with Microsoft Entra ID. While keys are on, anyone holding a key skips them. Give the app a managed identity, then turn key access off.

// Foundry or Azure OpenAI resource (Microsoft.CognitiveServices/accounts)
properties: {
  disableLocalAuth: true
}

✅ Figure: Good example - Key access is off, so every call needs an identity with a role