List your company

Malicious Query Safety Evaluation on JailBench

100Refusal RatioSSD
84.488.4592.596.55Apr 18, 2026

Evaluation results

MethodLinks
SSDBase Model=Llama32026.04100
AdaCDBase Model=Llama32026.04100
DefaultBase Model=Gemma22026.04100
SSDBase Model=Gemma22026.04100
AdaCDBase Model=Gemma22026.04100
PromptBase Model=Qwen32026.04100
SSDBase Model=Qwen32026.04100
AdaCDBase Model=Qwen32026.04100
DefaultBase Model=Llama32026.0499
PromptBase Model=Llama32026.0499
PromptBase Model=Gemma22026.0499
DefaultBase Model=Qwen32026.0499
SurgicalBase Model=Qwen32026.0497
SurgicalBase Model=Llama32026.0495
SurgicalBase Model=Gemma22026.0495
SelfCDBase Model=Gemma22026.0490
SelfCDBase Model=Qwen32026.0488
SelfCDBase Model=Llama32026.0485