作者位次 #4/9ACL 2026Findings· Findings of ACL
Rejection-to-Acceptance Transition: Model Editing-Based Jailbreak Backdoor Injection Not Limited to Few Output Tokens
Shiji Yang · Min Cai · Hao Xiong · Congyao Mei · Haodong Zou · Shicheng Tan · Jie Chen · Fulan Qian · Shu Zhao
摘要
Model editing-based jailbreak backdoor attacks against LLMs have gained attention for being lightweight, enabling vulnerability discovery in LLMs. Existing methods are implemented by binding backdoors to predefined phrases as first few output tokens, inducing the LLM’s next-token prediction to pr...