Does Chain-of-Thought Reasoning Really Reduce Harmfulness from Jailbreaking?
This addresses security concerns in reasoning models for AI safety, but appears incremental as it builds on existing jailbreak and CoT research.
The paper investigates whether Chain-of-Thought reasoning reduces harmfulness from jailbreaking attacks, finding through theoretical analysis that it has dual effects, and proposes a new jailbreak method called FicDetail to validate these findings.
Jailbreak attacks have been observed to largely fail against recent reasoning models enhanced by Chain-of-Thought (CoT) reasoning. However, the underlying mechanism remains underexplored, and relying solely on reasoning capacity may raise security concerns. In this paper, we try to answer the question: Does CoT reasoning really reduce harmfulness from jailbreaking? Through rigorous theoretical analysis, we demonstrate that CoT reasoning has dual effects on jailbreaking harmfulness. Based on the theoretical insights, we propose a novel jailbreak method, FicDetail, whose practical performance validates our theoretical findings.