Question
For each of the following statements, select Yes if the statement is true. Otherwise, select No.
NOTE: Each correct selection is worth one point.
Explanation
Statement 1 (Yes): A prompt injection attack tries to coerce the model into doing something it should not, so it can lead to unintended behavior and to sensitive data being revealed. This is why Copilot applies layered protections against these attacks throughout prompt processing.
Statement 2 (Yes): Malicious instructions can be typed directly into a prompt or embedded indirectly in content the model processes, such as text files, images, code, or encoded text. Indirect attacks work because the instructions are not in the prompt itself.
Statement 3 (Yes): Prompt injection targets the way generative AI systems interpret natural language, where instructions and data are not cleanly separated. Other injection attacks, such as SQL injection, exploit different mechanisms and are not the same vulnerability.
