Dawn Song

Can Editing LLMs Inject Harm?

能否通过编辑LLM注入危害？

Canyu Chen, Baixiang Huang, Zekun Li, Zhaorun Chen, Shiyang Lai, Xiongxiao Xu, Jia-Chen Gu, Jindong Gu, Huaxiu Yao, Chaowei Xiao, Xifeng Yan, William Yang Wang, Philip Torr, Dawn Song, Kai Shu

机构 * Northwestern University（西北大学） 纠错

AI总结 本文提出编辑攻击作为LLM安全威胁，揭示了通过编辑注入虚假信息和偏见的风险及高隐蔽性。

Comments Accepted to Proceedings of AAAI 2026. The first two authors contributed equally. 7 pages for main paper, 31 pages including appendix. The code, results, dataset for this paper and more resources are on the project website: https://llm-editing.github.io