Prompt Injection Drills

1. What is prompt injection primarily about?

Exploiting untrusted content as instructions Enhancing the performance of AI models Improving data retrieval methods Maintaining security in trusted systems

2. In the example of the email-summarizing assistant, what does the attacker do to manipulate the AI's actions?

Sends a phishing email Forwards the emails directly Injects commands into the email content Tricks the user into changing settings

3. Which of the following is considered the strongest defense against prompt injection?

Instruction hierarchy Privilege separation Output filtering Human confirmation

4. What is a key characteristic of the defenses against prompt injection discussed in the lesson?

They are all technology-based solutions They focus on training the AI model They resemble classic security principles They require user intervention for every action

Continue: Classical ML in One Lesson → Review this lesson Retake quiz