@article{llmsencodeharmfulnessandrefusal, title = {LLMs Encode Harmfulness and Refusal Separately}, author = {Jiachen Zhao and Jing Huang and Zhengxuan Wu and David Bau and Weiyan Shi}, year = {2025}, eprint = {2507.11878}, archivePrefix = {arXiv}, url = {https://arxiv.org/abs/2507.11878v1}, }