Privacy and security risks · Question 41 of 60
After an extremely long input, a chatbot outputs fragments that look like email addresses of other people. Which attack vector was probably used?
1 answer is correct. Hard Created with AI
Show answer
Answer
- D Context manipulation by exceeding the context window
Explanation
Very long inputs can overwhelm the context window. The model may then reproduce excerpts from training data, apparently personal details in this case.