AI and LLMs · Senior · Question 22 of 75
An AI feature starts producing harmful output for many users. What is the first move?
- 1Use a kill switch or flag to disable or degrade it, then investigate
- 2Wait for enough complaints to confirm the pattern is real
- 3Edit the prompt in production quickly without any testing
Show the answer
Correct answer: Use a kill switch or flag to disable or degrade it, then investigate
Explanation
AI features need the same incident tools as others: flags, rollbacks, fallbacks and a post-mortem that adds new eval cases.