Prompt Injection and Defenses

1. What is the main security risk associated with AI applications as stated in the lesson?

Data leakage Prompt injection Overfitting Model retraining

2. Which of the following defenses helps to keep model instructions separate from untrusted content?

Using large training datasets Implementing user feedback Separate instructions from content Increasing model complexity

3. According to the lesson, what should be done with model outputs before they are sent to a sink?

They should be logged They must be displayed immediately They need to be validated and escaped They should be stored for analysis

4. What does the lesson suggest about the confidence level of AI responses to hostile input?

Always trust the model Fail closed on low confidence Implement user intervention always Increase model complexity