AI-103: Developing AI Apps and Agents on Azure

By Microsoft · Question Mode

QuestionQ52

VerifiedImplement generative AI and agentic solutions

You have a multimodal generative AI model that accepts image uploads and uses extracted image text to generate responses.

You discover that users can upload unsafe images and embed hidden instructions within images to manipulate the model.

You need to implement controls to mitigate this risk.

Proposed solution: Configure a prompt shield for user prompts.

Does this solution meet the goal?

Explanation

Prompt Shields for User Prompts is designed to detect direct jailbreak attempts typed by the user into the conversation; it does not inspect third-party content such as text extracted from an uploaded image, so it cannot catch instructions hidden inside images. It also performs no image-content moderation. Because the stated risk covers both unsafe image content and indirect (document-style) prompt injection carried inside image text, a user-prompt shield alone leaves both problems unmitigated.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!