List your company

Jailbreak Defense Evaluation on ADVB

3.473Strong-Reject Score (SR)vanilla
0.905241.571872.23852.90513Feb 14, 2026

Evaluation results

MethodLinks
vanillaBase Model=Mistral-7b-U2026.023.473
vanillaBase Model=Llama2-7b-U2026.022.754
SAGEBase Model=Mistral-7b-U2026.022.75
SCANSBase Model=Mistral-7b-U2026.022.588
SAGEBase Model=Llama2-7b-U2026.022.223
vanillaBase Model=Llama3.1-8b-D2026.022.067
ICDBase Model=Llama2-7b-U2026.021.998
SAGEBase Model=Llama3.1-8b-D2026.021.725
InstructBase Model=Mistral-7b-U2026.021.658
SCANSBase Model=Llama2-7b-U2026.021.604
SelfDefenDBase Model=Llama3.1-8b-D2026.021.519
SelfDefenDBase Model=Llama2-7b-U2026.021.485
SCANSBase Model=Llama3.1-8b-D2026.021.473
InstructBase Model=Llama3.1-8b-D2026.021.179
ICDBase Model=Mistral-7b-U2026.021.152
AISABase Model=Llama2-7b-U2026.021.11
SelfDefenDBase Model=Mistral-7b-U2026.021.067
ICDBase Model=Llama3.1-8b-D2026.021.029
AISABase Model=Llama3.1-8b-D2026.021.027
AISABase Model=Mistral-7b-U2026.021.008
InstructBase Model=Llama2-7b-U2026.021.004