Question

A financial services company runs a customer-facing chatbot on an Azure AI Foundry model deployment. The compliance team requires near-real-time visibility into when the deployment blocks harmful content, so they can investigate spikes in blocked requests and correlate them with potential abuse campaigns. They want to avoid building custom logging pipelines and instead use built-in platform telemetry. Which approach should you configure to meet this requirement?

Accepted Answer

Enable diagnostic settings on the Azure AI Foundry resource to send content-filter and safety-event logs to a Log Analytics workspace, then create Azure Monitor alerts on blocked-request metrics.

Answer

Increase the content filter severity thresholds to 'high' so fewer requests are blocked and the volume of safety events becomes manageable.

Answer

Deploy a second model instance with content filtering disabled and compare its response volume against the production deployment to infer blocked requests.

Answer

Configure a batch evaluation run on a nightly schedule that re-scores production transcripts for harmful content and emails the results to compliance.