TY  - RPRT
TI  - STAR: Detecting Inference-time Backdoors in LLM Reasoning via State-Transition Amplification Ratio
AU  - Seong-Gyu Park
AU  - Sohee Park
AU  - Jisu Lee
AU  - Hyunsik Na
AU  - Daeseon Choi
PY  - 2026
UR  - https://arxiv.org/abs/2601.08511
ID  - 2601.08511
ER  -