A financial services company is deploying a generative AI application on Azure that exposes a custom model inference endpoint to internal business users. The security architect must design controls aligned with the Well-Architected Framework and Microsoft guidance for secure AI adoption. The primary concerns are protecting the model from prompt injection and jailbreak attempts, preventing sensitive data leakage in responses, and ensuring only authorized identities can call the endpoint. Which combination of controls should the architect prioritize to address these AI-specific threats at the application layer?