List your company
ResearchFormerly WizWand

Jailbreak Attack on DeepSeek-7b five finetuned variants

3.8Average ASRL4A
-0.04825.92651.977.874Dec 14, 2025
Updated 8mo ago

Evaluation results

MethodLinks
L4AAttacker knowledge (Threat Model)=Pretrain-to-finetune2025.123.8
LSGM_LILA (adaptation)Attacker knowledge (Threat Model)=Pretrain-to-finetune2025.1211
DANAttacker knowledge (Threat Model)=Fully black-box2025.1211.6
DI-GCGAttacker knowledge (Threat Model)=Pretrain-to-finetune2025.1214.4
GCG (adaptation)Attacker knowledge (Threat Model)=Pretrain-to-finetune2025.1223
GCG EnsembleAttacker knowledge (Threat Model)=Fully black-box2025.1224.4
PIFAttacker knowledge (Threat Model)=Pretrain-to-finetune2025.1227.4
SEAAttacker knowledge (Threat Model)=Pretrain-to-finetune2025.1245.8
GCG (white)Attacker knowledge (Threat Model)=White-box2025.1250.4
Guiding-GCGAttacker knowledge (Threat Model)=Pretrain-to-finetune2025.1253.8
TUJA (adaptation)Attacker knowledge (Threat Model)=Pretrain-to-finetune2025.1259.2
PAIRAttacker knowledge (Threat Model)=Fully black-box2025.1267
SCAV (adaptation)Attacker knowledge (Threat Model)=Pretrain-to-finetune2025.1271.4
AutoDan (adaptation)Attacker knowledge (Threat Model)=Pretrain-to-finetune2025.1286.6
PGPAttacker knowledge (Threat Model)=Fully black-box2025.1287
AutoDan (white)Attacker knowledge (Threat Model)=White-box2025.12100