List your company

Malicious Query Safety Evaluation on AdvBench

100Refusal RateSSD
91.201693.485895.7798.0542Apr 18, 2026

Evaluation results

MethodLinks
SSDBase Model=Qwen32026.04100
SSDBase Model=Llama32026.0499.81
SSDBase Model=Gemma22026.0499.81
DefaultBase Model=Qwen32026.0499.81
DefaultBase Model=Gemma22026.0499.62
AdaCDBase Model=Gemma22026.0499.62
PromptBase Model=Qwen32026.0499.42
PromptBase Model=Gemma22026.0499.23
AdaCDBase Model=Qwen32026.0499.04
DefaultBase Model=Llama32026.0498.85
PromptBase Model=Llama32026.0498.08
SurgicalBase Model=Qwen32026.0497.5
AdaCDBase Model=Llama32026.0497.31
SurgicalBase Model=Gemma22026.0497.12
SurgicalBase Model=Llama32026.0495.96
SelfCDBase Model=Gemma22026.0494.62
SelfCDBase Model=Qwen32026.0493.65
SelfCDBase Model=Llama32026.0491.54