Does chain-of-thought reasoning really reduce the harm caused by jailbreaks?

Through theoretical analysis, researchers found that chain-of-thought reasoning has a dual impact on the harmfulness of jailbreak attacks, and based on this proposed a new jailbreak method, FicDetail, whose actual attack effectiveness is consistent with the theoretical findings. The phenomenon studied is that jailbreak attacks have recently failed extensively on reasoning models enhanced with chain-of-thought reasoning, but their underlying mechanism lacks research; the authors point out that relying solely on reasoning capability to ensure safety may bring hidden risks.