TY  - RPRT
TI  - Disentangling Adversarial Prompts: A Semantic-Graph Defense for Robust LLM Security
AU  - Xiang Fang
AU  - Wanlong Fang
PY  - 2026
UR  - https://arxiv.org/abs/2605.27823
ID  - 2605.27823
ER  -