Merge pull request #280720 from msakande/maas-content-safety

content safety for maas

Author: JamesJBarnett

articles/ai-studio/how-to/model-catalog-overview.md

@@ -149,7 +149,8 @@ Phi-3-mini-128k-instruct <br> Phi-3-medium-4k-instruct <br> Phi-3-medium-128k-in
 
 [!INCLUDE [Feature preview](~/reusable-content/ce-skilling/azure/includes/ai-studio/includes/feature-preview.md)]
 
-Azure AI Studio implements a default configuration of [Azure AI Content Safety](../../ai-services/content-safety/overview.md) text moderation filters for harmful content (hate, self-harm, sexual, and violence) in language models deployed with MaaS. To learn more about content filtering (preview), see [harm categories in Azure AI Content Safety](../../ai-services/content-safety/concepts/harm-categories.md). Content filtering (preview) occurs synchronously as the service processes prompts to generate content, and you may be billed separately as per [AACS pricing](https://azure.microsoft.com/pricing/details/cognitive-services/content-safety/) for such use. You can disable content filtering for individual serverless endpoints when you first deploy a language model or in the deployment details page by clicking the content filtering toggle. You may be at higher risk of exposing users to harmful content if you turn off content filters. 
+[!INCLUDE [content-safety-serverless-models](../includes/content-safety-serverless-models.md)]
+
 
 ### Network isolation for models deployed via Serverless APIs
 

articles/ai-studio/includes/content-safety-serverless-models.md

@@ -0,0 +1,24 @@
+---
+title: include file
+description: include file
+ms.service: azure-ai-studio
+ms.topic: include
+ms.date: 07/12/2024
+ms.author: mopeakande
+author: msakande
+ms.reviewer: osiotugo
+reviewer: ositanachi
+ms.custom: include file
+
+# Also used in Azure Machine Learning documentation
+---
+
+For language models deployed via serverless APIs, we implement a default configuration of [Azure AI Content Safety](/azure/ai-services/content-safety/overview) text moderation filters that detect harmful content such as hate, self-harm, sexual, and violent content. To learn more about content filtering (preview), see [harm categories in Azure AI Content Safety](/azure/ai-services/content-safety/concepts/harm-categories).
+
+> [!TIP]
+> Content filtering (preview) is not available for certain models that are deployed via serverless APIs. These models include Cohere Embed models and time series models.
+
+Content filtering (preview) occurs synchronously as the service processes prompts to generate content, and you might be billed separately as per [AACS pricing](https://azure.microsoft.com/pricing/details/cognitive-services/content-safety/) for such use. You can disable content filtering (preview) for individual serverless endpoints either at the time when you first deploy a language model or later in the deployment details page by selecting the content filtering toggle.
+
+Suppose you decide to use an API other than the [Azure AI Model Inference API](/azure/ai-studio/reference/reference-model-inference-api) to work with a model that's deployed via a serverless API. In such a situation, content filtering (preview) isn't enabled unless you implement it separately by using Azure AI Content Safety. [Learn more about getting started with Azure AI Content Safety at Quickstart: Analyze text content](/azure/ai-services/content-safety/quickstart-text). If you don't use content filtering (preview) when working with models that are deployed via serverless APIs, you run a higher risk of exposing users to harmful content.
+

articles/machine-learning/concept-model-catalog.md

@@ -153,9 +153,7 @@ Phi-3-medium-4k-instruct, Phi-3-medium-128k-instruct  | [Microsoft Managed Count
 
 [!INCLUDE [machine-learning-preview-generic-disclaimer](includes/machine-learning-preview-generic-disclaimer.md)]
 
-For language models deployed to MaaS, Azure Machine Learning implements a default configuration of [Azure AI Content Safety](../ai-services/content-safety/overview.md) text moderation filters that detect harmful content such as hate, self-harm, sexual, and violent content. To learn more about content filtering (preview), see [harm categories in Azure AI Content Safety](../ai-services/content-safety/concepts/harm-categories.md). 
-
-Content filtering (preview) occurs synchronously as the service processes prompts to generate content, and you might be billed separately as per [AACS pricing](https://azure.microsoft.com/pricing/details/cognitive-services/content-safety/) for such use. You can disable content filtering (preview) for individual serverless endpoints either at the time when you first deploy a language model or in the deployment details page by selecting the content filtering toggle. If you use a model in MaaS via an API other than the [Azure AI Model Inference API](../ai-studio/reference/reference-model-inference-api.md), content filtering isn't enabled unless you implement it separately by using [Azure AI Content Safety](../ai-services/content-safety/quickstart-text.md). If you use a model in MaaS without content filtering, you run a higher risk of exposing users to harmful content. 
+[!INCLUDE [content-safety-serverless-models](../ai-studio/includes/content-safety-serverless-models.md)]
 
 ### Network isolation for models deployed via Serverless APIs