TY  - RPRT
TI  - Addition by Subtraction: Toward Mitigating Fragile Safety Alignment in Large Language Models
AU  - Hamid Dashtbani
AU  - Alexandra Dmitrienko
PY  - 2026
UR  - https://arxiv.org/abs/2611.06553
ID  - 2611.06553
ER  -