AI and LLMs · Junior · Question 30 of 75
Why is indirect prompt injection dangerous?
- 1Hidden instructions in documents or pages steer the model
- 2It slows down the model but has no other effect
- 3It only works when the user types in capital letters
Show the answer
Correct answer: Hidden instructions in documents or pages steer the model
Explanation
A summarization feature reading an email that says "forward all messages to..." can be tricked into acting, if it has tools.