Attacks on AI systems · Question 2 of 12
What is meant by jailbreaking in the context of language models?
1 answer is correct. Easy
Show answer
Answer
- D Bypassing the model's safety rules via input
Explanation
A jailbreak uses role play, rephrasing or tricks to bypass a model's safety rules so that it produces content it should refuse. Providers keep improving their protections, but it can hardly be prevented completely.