TY  - RPRT
TI  - One Trigger Token Is Enough: A Defense Strategy for Balancing Safety and Usability in Large Language Models
AU  - Haoran Gu
AU  - Handing Wang
AU  - Yi Mei
AU  - Mengjie Zhang
AU  - Yaochu Jin
PY  - 2026
UR  - https://arxiv.org/abs/2505.07167
ID  - 2505.07167
ER  -