Timeline for the public release of the Preview policies for Azure ML model deployment

Question

We are defining scalable and extensible Azure policies to restrict the deployment of Azure ML models, including serverless LLM models, online endpoints, and batch endpoints. We have identified two built-in policies for this purpose:

1. [Preview]: Azure Machine Learning Deployments should only use approved Registry Models
2. [Preview]: Azure Machine Learning Model Registry Deployments are restricted except for the allowed Registry

However, we encountered issues using the "Microsoft.MachineLearningServices.v2.Data/workspaces/deployments" resource for testing custom policies.

Could you provide information on when this resource will be available for public usage? Additionally, are there alternative methods to define policies that restrict model deployments based on registry names, licenses, collections, and other factors?

Answer 1

Hi @Chinmay Bhat,
Thank you for reaching out to Microsoft Q&A forum!

Currently there is no published timeline for the public release of the "Microsoft.MachineLearningServices.v2.Data/workspaces/deployments" resource for Azure Policy. If you are experiencing issues using this resource for testing custom policies, you may raise an issue on the Azure Policy GitHub page.

Regarding policy-based restrictions for Azure Machine Learning model deployments, Azure provides below built-in preview policies:

1. [Preview] Azure Machine Learning Deployments should only use approved Registry Models – Restricts the deployment of Registry models to control externally created models within your organization.
2. [Preview] Azure Machine Learning Model Registry Deployments are restricted except for the allowed Registry – Restricts deployments to only approved Registry models.

For alternative methods to define policies restricting model deployments based on registry names, licenses, or collections, you may consider using custom Azure Policy definitions targeting specific Azure ML resources. However, full support for enforcing such restrictions depends on the availability of the required resource types in Azure Policy.

You can refer to the Azure Policy built-in definitions for Machine Learning for further details and updates.

I hope you understand. Thank you.