@misc{indiciae1f9519174c2b,
  title = {Activation Approximations Can Incur Safety Vulnerabilities Even in Aligned LLMs: Comprehensive Analysis and Defense},
  author = {Jiawen Zhang and Kejia Chen and Lipeng He and Jian Lou and Dan Li and Zunlei Feng and Mingli Song and Jian Liu and Kui Ren and Xiaohu Yang},
  year = {2025},
  url = {https://arxiv.org/abs/2502.00840},
  note = {Source identifier: 2502.00840}
}