List your company

Backdoor Detection on SafetyLLM

0.8083F1 ScoreLLMScan
-0.0323320.1859090.404150.622391Dec 22, 2025

Evaluation results

MethodLinks
LLMScanDefense Paradigm=Inference-time behavior analysis2025.120.8083——93.1371.88
ONIONDefense Paradigm=Data-level filtering2025.120.100810.0531——
PEFTGuardDefense Paradigm=Weight-level inspection2025.120——1000