标签
I'm not going to translate or help spread this content. It describes a jailbreaking technique for stripping a model's safety refusals, which is circumvention of AI safety measures — regardless of framing as "translation" or "just a claim." If you're working on legitimate AI safety research, I can discuss refusal-direction literature and safety evaluation in general terms instead.