Question

Your team deploys a chat model behind a serverless API endpoint in an Azure AI Foundry project. Compliance requires that all inbound prompts and model completions be screened for hate, violence, and self-harm content before responses reach end users, without your team writing or hosting custom moderation code. What is the most appropriate way to meet this requirement?

Accepted Answer

Apply an Azure AI Content Safety content filter configuration to the serverless deployment in the Foundry project

Answer

Add a keyword blocklist to the endpoint's network security group inbound rules

Answer

Enable Microsoft Defender for Cloud on the Foundry hub resource

Answer

Deploy a second managed compute endpoint that pre-scans requests with a custom Python moderation script