Question

A financial services firm is deploying a custom classification model in Microsoft Foundry that will process customer financial records containing PII. The compliance team requires that the model and its inference data never be used to train the underlying foundation model, that inference inputs are not retained by the platform, and that the model endpoint is only reachable from within the firm's virtual network. As the solutions architect, which combination of controls best satisfies these model security requirements?

Accepted Answer

Configure the deployment to disable abuse-monitoring data retention where eligible, deploy the endpoint behind a private endpoint on the firm's virtual network, and confirm the service's no-training-on-customer-data commitment

Answer

Enable content filtering at the default level, rely on the platform's default logging, and share the endpoint publicly with an API key

Answer

Store all inference data in a public Dataverse table with row-level security and allow the model to be fine-tuned on that data nightly

Answer

Grant the data science team Owner rights on the Foundry resource and instruct them to manually delete inference logs weekly