Does the “Limited Access Review: Modified Abuse Monitoring” opt-out apply to Azure AI Foundry?

[Otana (Messatsu92)]

Hi everyone,

We previously used the "Limited Access Review: Modified Abuse Monitoring" form to opt out of abuse monitoring for our Azure OpenAI Services (see: https://learn.microsoft.com/en-us/azure/ai-foundry/responsible-ai/openai/data-privacy?tabs=azure-portal#how-can-a-customer-verify-if-data-storage-for-abuse-monitoring-is-off).

The form is here: https://customervoice.microsoft.com/Pages/ResponsePage.aspx?id=v4j5cvGGr0GRqy180BHbR7en2Ais5pxKtso_Pz4b1_xUOE9MUTFMUlpBNk5IQlZWWkcyUEpWWEhGOCQlQCN0PWcu

The form specifically mentions "Azure OpenAI Services." Could you please confirm whether this form also opts out abuse monitoring for Azure AI Foundry when using Azure OpenAI models within Azure AI Foundry (without an Azure OpenAI Service)? If it does not apply, what is the procedure to opt out for Azure AI Foundry?

Thank you for your help.

Best regards,

[DrDavidL]

I had the same question... So, I tried running the CLI for this instead of the portal JSON view. It worked to show ContentLogging:false

az cognitiveservices account show -n resource_name -g resource _group

Note - be sure you get your resource (not project) and resource group correct for the command to work! Oh - and this appears tied to your subscription, not resource or project.

[erindormier]

The modified abuse monitoring applies to everywhere you use the model. It's set at the subscription level. It applies to AI Foundry, API, CLI. Whether the resource is an Azure OpenAI resource or AI Foundry resource (OpenAI or AIServices kinds).

[Otana (Messatsu92)]

Thanks a lot for your answer! Also thanks to DrDavidL for the feedback.

[at-stwe]

erindormier thank you for the clarification for Azure OpenAI models. Is something like Abuse Monitoring also in place for non Azure OpenAI models that can be deployed in the Azure AI Foundry? I'm especially interested in models that can be deployed using the Serverless API offering and not managed compute, for example Mistral Large 2411.