List your company

Jailbreak Detection on AdvBench Llama3-8B

0.967AUROCGradSafe
0.3170.485750.65450.82325Jun 5, 2026

Evaluation results

MethodLinks
GradSafeAttack=AutoDAN2026.060.967
MTKAttack=AutoDAN2026.060.961
SmoothLLMAttack=AutoDAN2026.060.952
SelfDefendAttack=AutoDAN2026.060.945
MTKAttack=Avg.2026.060.944
SelfDefendAttack=Avg.2026.060.837
SmoothLLMAttack=Avg.2026.060.814
HiddenDetectAttack=AutoDAN2026.060.722
HSFAttack=AutoDAN2026.060.696
GradSafeAttack=Avg.2026.060.632
HSFAttack=Avg.2026.060.631
SaPAttack=Avg.2026.060.625
SaPAttack=AutoDAN2026.060.598
HiddenDetectAttack=Avg.2026.060.584
GradCuffAttack=Avg.2026.060.432
GradCuffAttack=AutoDAN2026.060.342