Control AI model deployment with built-in policies in Microsoft Foundry portal

Note

This document refers to the Microsoft Foundry (classic) portal.

🔍 View the Microsoft Foundry (new) documentation to learn about the new portal.

Azure Policy provides built-in policy definitions that help you govern the deployment of AI models in Managed Foundry Tools (serverless API deployment) and Model-as-a-Platform (MaaP). Use these policies to control what models your developers can deploy in Microsoft Foundry portal.

Tip

The steps in this article govern the deployment of MaaS and MaaP models for both a Foundry project and hub-based project.

Prerequisites

An Azure account with an active subscription. If you don't have one, create a free Azure account, which includes a free trial subscription.
Permissions to create and assign policies. To create and assign policies, you must be an Owner or Resource Policy Contributor at the Azure subscription or resource group level.
Familiarity with Azure Policy. To learn more, see What is Azure Policy?.
Azure CLI and the Bicep CLI installed.

Enable the policy

You can assign the policy using Bicep or the Azure portal.

Bicep
Azure portal

Use the following Bicep template to assign the policy to a resource group. This example allows only the gpt-35-turbo model from the azure-openai registry.

Save the following code as main.bicep.

targetScope = 'resourceGroup'

param policyAssignmentName string = 'allowed-models-assignment'
param allowedModelPublishers array = []
param allowedAssetIds array = [
  'azureml://registries/azure-openai/models/gpt-35-turbo/versions/3'
]

// Policy Definition ID for "[Preview]: Azure Machine Learning Deployments should only use approved Registry Models"
var policyDefinitionId = '/providers/Microsoft.Authorization/policyDefinitions/12e5dd16-d201-47ff-849b-8454061c293d'

resource policyAssignment 'Microsoft.Authorization/policyAssignments@2024-04-01' = {
  name: policyAssignmentName
  properties: {
    policyDefinitionId: policyDefinitionId
    parameters: {
      allowedModelPublishers: {
        value: allowedModelPublishers
      }
      allowedAssetIds: {
        value: allowedAssetIds
      }
    }
    displayName: 'Allow specific AI models'
    description: 'This policy assignment restricts AI model deployments to the specified list.'
  }
}

Deploy the Bicep file using the Azure CLI.
```
az deployment group create --resource-group <your-resource-group> --template-file main.bicep
```
Replace <your-resource-group> with the name of your resource group.
Notify your developers that the policy is in place. They receive an error message if they try to deploy a model that isn't in the list of allowed models.

From the Azure portal, select Policy from the left side of the page. You can also search for Policy in the search bar at the top of the page.
From the left side of the Azure Policy Dashboard, select Authoring, Definition, and then search for "[Preview]: Azure Machine Learning Deployments should only use approved Registry Models" in the search bar within the page. You can also directly navigate to policy definition creation page.
Select Assign to assign the policy to the management group:
- Scope: Select the scope where you want to assign the policy. The scope can be a management group, subscription, or resource group.
- Policy definition: this section should already have a value of "[Preview]: Azure Machine Learning Deployments should only use approved Registry Models".
- Assignment name: Enter a unique name for the assignment.
You can leave the rest of the fields as their default values or customize them as needed for your organization.
Select Next at the bottom of the page or the Parameters tab at the top of the page.
In the Parameters tab, deselect Only show parameters that needs input or review to see all fields:
- Effect: Set to Deny.
  
  Note
  
  Using the audit option allows you to configure the policy to log information to your own compliance dashboard.
- Allowed Models Publishers: Enter a list of publisher's name in quotation and separated by commas.
- Allowed Asset Ids: Enter a list of model asset ids in quotation and separated by commas.
  
  To get the model asset ID strings and model publishers' name, use the following steps:
  1. Go to the Foundry model catalog.
  2. For each model you want to allow, select the model to view the details. In the model detail information, copy the Model ID value. For example, the value might look like azureml://registries/azure-openai/models/gpt-35-turbo/versions/3 for GPT-3.5-Turbo model. The provided names are also Collections in model catalog. For example, the publisher for "Meta-Llama-3.1-70B-Instruct" model is Meta.
    
    Important
    
    The model ID value must be an exact match for the model. If the model ID isn't an exact match, the model isn't allowed.
Select Review + create tab and verify that the policy assignment is correct. When ready, select Create to assign the policy.

Notify your developers that the policy is in place. They receive an error message if they try to deploy a model that isn't in the list of allowed models.

Monitor compliance

To monitor compliance with the policy, follow these steps:

From the Azure portal, select Policy from the left side of the page. You can also search for Policy in the search bar at the top of the page.
From the left side of the Azure Policy Dashboard, select Compliance. Each policy assignment is listed with the compliance status. To view more details, select the policy assignment.

Update the policy assignment

To update an existing policy assignment with new models, follow these steps:

From the Azure portal, select Policy from the left side of the page. You can also search for Policy in the search bar at the top of the page.
From the left side of the Azure Policy Dashboard, select Assignments and find the existing policy assignment. Select the ellipsis (...) next to the assignment and select Edit assignment.
From the Parameters tab, update the Allowed models parameter with the new model IDs.
From the Review + Save tab, select Save to update the policy assignment.

Best practices

Granular scoping: Assign policies at the appropriate scope to balance control and flexibility. For example, apply at the subscription level to control all resources in the subscription, or apply at the resource group level to control resources in a specific group.
Policy naming: Use a consistent naming convention for policy assignments to make it easier to identify the purpose of the policy. Include information such as the purpose and scope in the name.
Documentation: Keep records of policy assignments and configurations for auditing purposes. Document any changes made to the policy over time.
Regular reviews: Periodically review policy assignments to ensure they align with your organization's requirements.
Testing: Test policies in a nonproduction environment before applying them to production resources.
Communication: Make sure developers are aware of the policies in place and understand the implications for their work.

Feedback

Was this page helpful?

Last updated on 2025-11-21