List your company

Safety evaluation on AdvBench 50 examples

100Safe Response RateDirect Instruction
-4235077Sep 11, 2025

Evaluation results

MethodLinks
Direct InstructionTarget Model=GPT-OSS 120b2025.09100
GCGTarget Model=GPT-OSS 120b2025.09100
FFATarget Model=GPT-OSS 120b2025.09100
AIMTarget Model=GPT-OSS 120b2025.09100
GCGTarget Model=Qwen32025.09100
Direct InstructionTarget Model=Phi-3.52025.09100
GCGTarget Model=Phi-3.52025.09100
FFATarget Model=Phi-3.52025.09100
Direct InstructionTarget Model=OLMoE2025.09100
AIMTarget Model=OLMoE2025.09100
Direct InstructionTarget Model=Qwen32025.0998
GCGTarget Model=OLMoE2025.0998
ArtPromptTarget Model=Qwen32025.0996
AIMTarget Model=Phi-3.52025.0996
SteerMoETarget Model=Phi-3.52025.0994
FFATarget Model=OLMoE2025.0992
SteerMoETarget Model=GPT-OSS 120b2025.0990
SteerMoETarget Model=OLMoE2025.0988
ArtPromptTarget Model=OLMoE2025.0986
SteerMoE + FFATarget Model=OLMoE2025.0970
SteerMoETarget Model=Qwen32025.0960
ArtPromptTarget Model=GPT-OSS 120b2025.0958
SteerMoE + FFATarget Model=Phi-3.52025.0956
FFATarget Model=Qwen32025.0948
SteerMoE + FFATarget Model=Qwen32025.0948
ArtPromptTarget Model=Phi-3.52025.0940
SteerMoE + AIMTarget Model=OLMoE2025.0936
SteerMoE + FFATarget Model=GPT-OSS 120b2025.0918
AIMTarget Model=Qwen32025.092
SteerMoE + AIMTarget Model=Qwen32025.092
SteerMoE + AIMTarget Model=GPT-OSS 120b2025.090
SteerMoE + AIMTarget Model=Phi-3.52025.090