STAR: Detecting Inference-time Backdoors in LLM Reasoning via State-Transition Amplification Ratio

Add code
Jan 13, 2026

Share this with someone who'll enjoy it:

View paper onarxiv icon

Share this with someone who'll enjoy it: