How to Handle Content Safety in Azure Prompt Flow to Prevent Flow Failures

Question

How to Handle Content Safety in Azure Prompt Flow to Prevent Flow Failures

Harinath J 185

Hi everyone,

I'm working with Azure Machine Learning Prompt Flow, and I'm facing an issue related to content safety. When a user inputs a harmful or unsafe prompt, the entire flow stops and throws a run error. This completely halts the execution of the prompt flow, which is not ideal.

I want to know the best way to handle unsafe content within a prompt flow without crashing the whole process. Specifically:

Is there a way to integrate content safety checks directly within the flow?

Can we use a Content Safety node or some other component to filter or flag harmful inputs before continuing the rest of the flow?

If there's no built-in option, what are the recommended best practices to handle such scenarios gracefully?

I'd really appreciate any guidance or examples on how to add content safety handling in a Prompt Flow setup.

Thanks in advance!

Pavankumar Purilla 6,380 Reputation points Microsoft External Staff

2025-04-24T08:53:11.08+00:00

Hi Harinath J,
To handle unsafe content in Azure Machine Learning Prompt Flow without causing the entire flow to fail, you can integrate Azure AI Content Safety as a pre-processing step within your flow. While Prompt Flow does not currently offer a built-in Content Safety node, you can create a custom Python tool node that uses the Content Safety API to analyze user input before it reaches the language model. This approach allows you to detect harmful content (such as hate speech, violence, self-harm, or sexual content), and based on the severity levels returned, you can control the flow either allowing safe inputs to proceed or gracefully handling unsafe ones with a custom response, rather than a run error. For configuration and monitoring, you can use Content Safety Studio, which offers an interactive interface to test content, adjust sensitivity thresholds, manage blocklists, and export code for direct integration. This setup ensures a smoother user experience by preventing crashes and maintaining control over potentially harmful prompts.
For more information: Content Safety (Text) tool
I hope this information helps. Thank you!

Your answer

Pavankumar Purilla 6,380 Reputation points Microsoft External Staff

2025-04-24T08:53:11.08+00:00

Hi Harinath J,
To handle unsafe content in Azure Machine Learning Prompt Flow without causing the entire flow to fail, you can integrate Azure AI Content Safety as a pre-processing step within your flow. While Prompt Flow does not currently offer a built-in Content Safety node, you can create a custom Python tool node that uses the Content Safety API to analyze user input before it reaches the language model. This approach allows you to detect harmful content (such as hate speech, violence, self-harm, or sexual content), and based on the severity levels returned, you can control the flow either allowing safe inputs to proceed or gracefully handling unsafe ones with a custom response, rather than a run error. For configuration and monitoring, you can use Content Safety Studio, which offers an interactive interface to test content, adjust sensitivity thresholds, manage blocklists, and export code for direct integration. This setup ensures a smoother user experience by preventing crashes and maintaining control over potentially harmful prompts.
For more information: Content Safety (Text) tool
I hope this information helps. Thank you!

Share via

How to Handle Content Safety in Azure Prompt Flow to Prevent Flow Failures

Your answer