Correct:
* You configure a prompt shield for documents.
Prompt Shield for Documents: Highly Effective (Critical Defense)
How it helps: This shield specifically scans untrusted, third-party data inputs (like external documents or text extracted from uploaded images).
Mechanism: It evaluates the extracted image text before it is sent to the LLM to identify hidden jail
* You configure a prompt shield for user prompts.
Prompt Shield for User Prompts: Partially Effective (Defense in Depth)
How it helps: This shield targets direct jailbreak attempts written manually by the user in the text prompt field accompanying the upload.
Mechanism: It prevents the user from typing supporting instructions that prime the model to execute the hidden instructions found within the image.
* You configure image moderation to block unsafe content before processing the images.
Implementing rigorous image moderation is one of the most effective ways to secure multimodal AI systems against these threats. Moderation acts as a necessary gatekeeper, preventing malicious inputs from ever reaching the generative model.
Incorrect:
* You configure protected material detection.
Protected Material Detection: Ineffective for this Threat
Why it does not help: This feature is designed to scan model outputs to prevent the generation of copyrighted text, proprietary source code, or licensed imagery.
Limitation: It does not scan inputs for adversarial instructions and will not prevent a user from manipulating the model's logic.
Reference:
https://www.upgrad.com/blog/what-is-multimodal-ai/
https://learn.microsoft.com/en-us/azure/ai-services/content-safety/concepts/jailbreak-detection

By adding a comment like // what does function F do? to the prompt, you are instructing the GPT model to provide an explanation of the function's behavior. This approach directly asks the model to explain the code, which will help generate an explanation without requiring changes to the model or additional resource-intensive configurations.

Explanation:
* Orchestration pattern: The sequential template that passes outputs node-by-node
* Approval checkpoints: Add a condition statement
Use the sequential orchestration template because the agents must execute in a fixed, predictable order.
PolicyWriter first generates the proposed policy update, and its output is then passed directly to RiskReviewer. Microsoft defines the sequential pattern as passing the result from one agent to the next in a defined order, making it appropriate for deterministic, multistage processing. The group-chat pattern would allow control to move dynamically between agents, which would reduce execution predictability.
After RiskReviewer returns a structured risk classification, add a condition statement that evaluates the result. The low-risk branch can proceed directly to the finalization action without requesting human input. A separate higher-risk branch can route to an approval step, such as an Ask a question node. Adding an Ask a question node without conditional branching would pause every workflow execution, including low-risk updates, and therefore would not satisfy the automation requirement. Microsoft Foundry workflows support if
/else branching and condition expressions for selecting the next action.
Study Guide alignment: implement orchestrated multi-agent solutions and build autonomous or semiautonomous workflows with safeguards and approval-flow controls .

The solution does not meet the goal. Protected material detection is intended to identify large language model output that matches known protected text or code, such as copyrighted text, selected web content, song lyrics, articles, recipes, or code. Microsoft describes protected material detection as a control for preventing AI- generated content from reproducing known protected material, not as a control for image safety or prompt injection.
The stated risk has two parts: users can upload unsafe images, and users can embed hidden instructions in images to manipulate the model. Unsafe image uploads require image moderation, because Azure AI Content Safety provides image APIs that detect harmful content across modalities and can support blocking decisions by harm category and severity. Hidden instructions extracted from images are indirect prompt injection or document attacks; Microsoft Prompt Shields are the capability designed to detect user prompt attacks and document attacks, including harmful instructions embedded in third-party content.
Therefore, protected material detection alone does not mitigate either primary risk. Reference topics: Azure AI Content Safety, image moderation, Prompt Shields, document attacks, indirect prompt injection, and protected material detection.

The correct configuration choice is to set project-endpoint to the endpoint of the project.
azure-ai-project-endpoint: This parameter is required and must be set to the URL endpoint of your Microsoft Foundry project. This aligns with setting project-endpoint.
Reference:
https://learn.microsoft.com/en-us/azure/foundry/how-to/evaluation-github-action