TY  - RPRT
TI  - Understanding and Improving Continuous Adversarial Training for LLMs via In-context Learning Theory
AU  - Shaopeng Fu
AU  - Di Wang
PY  - 2026
UR  - https://arxiv.org/abs/2604.12817
ID  - 2604.12817
ER  -