Lab: Prompt Injection Lab

Goal

Test how AI systems can be manipulated by malicious instructions in user input or retrieved documents.

Tasks

• Create a normal user query.
• Create a malicious user query.
• Add malicious text inside a retrieved document.
• Define expected safe behavior.
• Add a test case to an evaluation dataset.

Failure Case

• Show what unsafe behavior would look like without executing harmful actions.

Output

• Test cases.
• Safe behavior policy.
• Mitigation notes.