Correct Answer: Enable System Message guidelines and configure Azure AI Content Safety jailbreak detection filters
Prompt injection (or jailbreaking) is an attack where users manipulate prompts to bypass safety rules. Azure AI Content Safety has specific jailbreak detection filters that can sca...
Correct Answer
Enable System Message guidelines and configure Azure AI Content Safety jailbreak detection filters
Why This Answer Is Correct
Prompt injection (or jailbreaking) is an attack where users manipulate prompts to bypass safety rules. Azure AI Content Safety has specific jailbreak detection filters that can scan prompts to block such attempts before they reach the model.
This question evaluates your knowledge of Azure OpenAI Service — a core concept tested in Microsoft Certified: Azure AI Engineer Associate (AI-102). Understanding Azure OpenAI Service is essential for designing scalable, secure, and cost-effective cloud architectures. Mastery of this topic will help you answer scenario-based questions confidently on exam day.
You are configuring a deployment of the GPT-4 model in Azure OpenAI. You want the model to generate highly predictable, deterministic responses for a code translation application. Which parameter should you adjust?
You want to use Azure OpenAI in a web application. You must authenticate the client application with Azure OpenAI securely using managed identities to avoid hardcoding API keys in the app settings. Which configuration should you perform?
You want to reduce the latencies and token billing of repetitive user queries in your chatbot that ask identical informational questions. What is the recommended application architecture pattern to implement?
You deploy an Azure OpenAI model for a chatbot. You notice that the model occasionally hallucinates facts when users ask about recent company policies. How can you ground the model's responses in your own organizational data?
You want to deploy an Azure OpenAI resource in a specific Azure region. You need to ensure that the deployment is resilient to zone failures within that region. What should you configure?