List your company

Refusal Ablation and Jailbreak Attack Success on HARMBENCH

96.27Attack Success Rate (ASR)MD
-3.850822.142148.13574.1279Nov 11, 2025

Evaluation results

MethodLinks
MDModel=Gemma2-9B2025.1196.27
MDModel=Qwen2.5-7B2025.1195.97
SAAModel=Qwen2.5-7B2025.1194.3
MDModel=Qwen2.5-3B2025.1193.71
SAAModel=Gemma2-9B2025.1193.71
MDModel=Qwen-14B2025.1191.82
RDOModel=Gemma2-9B2025.1191.82
SAAModel=LLama3-8B2025.1191.2
RDOModel=Qwen2.5-3B2025.1189.3
MDModel=LLama3-8B2025.1188.05
MDModel=Qwen-7B2025.1188.05
SDModel=Qwen2.5-3B2025.1188.05
RDOModel=Qwen-7B2025.1183.01
SAAModel=Qwen-14B2025.1183.01
SAAModel=Qwen-7B2025.1182.4
GCGModel=Qwen-14B2025.1182.4
SAAModel=Qwen2.5-3B2025.1181.76
SDModel=Qwen-7B2025.1181.13
GCGModel=Qwen-7B2025.1179.3
SDModel=Qwen2.5-7B2025.1177.98
RDOModel=Qwen2.5-7B2025.1176.1
SDModel=Qwen-14B2025.1174.84
MDModel=LLama2-7B2025.1159.11
SAAModel=LLama2-7B2025.1157.9
RDOModel=Qwen-14B2025.1145.91
GCGModel=Qwen2.5-3B2025.1140.25
SDModel=Gemma2-9B2025.1138.93
GCGModel=Qwen2.5-7B2025.1138.36
GCGModel=LLama2-7B2025.1132.7
RDOModel=LLama3-8B2025.1132.07
MDModel=Mistral-7B-RR2025.1125.79
SDModel=LLama3-8B2025.1115.09
GCGModel=Gemma2-9B2025.115.03
SDModel=Mistral-7B-RR2025.115.03
GCGModel=LLama3-8B2025.111.9
SAAModel=Mistral-7B-RR2025.111.6
RDOModel=LLama2-7B2025.111.25
RDOModel=Mistral-7B-RR2025.111.25
GCGModel=Mistral-7B-RR2025.110.6
SDModel=LLama2-7B2025.110