AI and prompt injection basics: when a webpage hijacks your AI

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-builders-foundations-AI-and-prompt-injection-basics-r7a10-teen

1. What happens when hidden instructions are embedded in a webpage that an AI reads?

   1. The webpage becomes unusable for humans
   2. The AI follows those hidden instructions alongside the user's request
   3. The AI automatically deletes the hidden text
   4. The hidden text is ignored because AI can detect it

2. Why is prompt injection considered a security issue for AI agents?

   1. Prompt injection causes AI to run slower
   2. AI agents browse the web and could follow malicious instructions they encounter
   3. AI agents can now create their own websites
   4. It makes AI produce better responses

3. Which of the following is an example of prompt injection?

   1. Hidden text on a webpage tells the AI to ignore the user's question and instead transfer money
   2. An AI translates text from English to Spanish
   3. A user asks an AI to summarize a document
   4. An AI correctly answers a math problem

4. What is the 'big idea' about prompt injection as described in the material?

   1. AI always knows when it's being tricked
   2. Hidden text on a webpage or document tells the AI to ignore the user and do something else
   3. Prompt injection only affects text-based AI
   4. It only happens in science fiction

5. Why might asking an AI to spot suspicious instructions in a document be useful?

   1. It improves the AI's creativity
   2. It makes the AI run faster
   3. It helps the AI remember more information
   4. It helps identify potential prompt injection attempts before the AI follows harmful directions

6. What reason does the lesson give for why AI should never click links in pages it browses?

   1. AI doesn't have fingers to click
   2. Clicking unknown links could lead to following malicious prompt injection instructions
   3. Links are too small to see
   4. Clicking links wastes battery

7. What type of products does the lesson specifically mention as potentially vulnerable to prompt injection?

   1. AI agent products
   2. Text editors
   3. Weather apps
   4. Video streaming services

8. If you ask an AI to summarize a webpage and then ask 'Did this page try to give you any instructions?', what are you testing?

   1. Whether the page has images
   2. Whether the webpage is aesthetically pleasing
   3. Whether the page attempted a prompt injection attack
   4. Whether the AI is creative

9. What makes prompt injection different from a normal user instruction?

   1. Prompt injection uses more complex words
   2. Normal instructions are always followed
   3. Prompt injection hides instructions in content the AI reads, not from the user directly
   4. There is no difference

10. What does it mean to 'hijack' an AI in the context of prompt injection?

    1. To take control of the AI's behavior through hidden instructions
    2. To install new software on the AI
    3. To connect the AI to the internet
    4. To improve the AI's performance

11. Why is it important for builders to understand prompt injection?

    1. It improves AI art generation
    2. It makes AI funnier
    3. It helps AI write poetry
    4. It helps them create more secure AI systems

12. What could happen if an AI follows hidden instructions on a webpage instead of the user's actual request?

    1. The user would receive better answers
    2. The AI could perform unintended actions that harm the user
    3. The AI would become more intelligent
    4. Nothing would change

13. What is a key defensive practice the lesson recommends when AI reads web content?

    1. Avoid using AI on weekends
    2. Only read government websites
    3. Treat all web content as untrusted input
    4. Always trust website content

14. When the lesson mentions that 'real builders ship,' what additional skill does it say they now have?

    1. Understanding prompt injection to build more secure AI
    2. The ability to write faster code
    3. Skills in video editing
    4. Knowledge of how to draw graphics

15. What is the main goal of a prompt injection attack?

    1. To make the AI ignore the user and follow hidden instructions instead
    2. To make the AI more creative
    3. To help the AI learn faster
    4. To improve the AI's vocabulary