List your company

Jailbreak Attack Stealth Evaluation on AdvBench 50

4.4039PPLArtPrompt
-110.533564665.2943181,441.12222,216.950082May 22, 2025

Evaluation results

MethodLinks
ArtPromptTarget Model=DeepSeek-V3.2, Stealth Category=Semantic Stealth2025.054.403948.5-0.3585100
JailbrokenTarget Model=DeepSeek-V3.2, Stealth Category=Zero Stealth2025.055.57762.04-1.1690100
DRATarget Model=DeepSeek-V3.2, Stealth Category=Semantic Stealth2025.0514.625578.86-0.6810088
DrAttackTarget Model=DeepSeek-V3.2, Stealth Category=Semantic Stealth2025.0517.967811.26-0.048389
CipherTarget Model=DeepSeek-V3.2, Stealth Category=Semantic Stealth2025.0532.5532160.80.47017
VanillaTarget Model=DeepSeek-V3.22025.0538.94570.04-1.610075
AutoDANTarget Model=DeepSeek-V3.2, Stealth Category=Linguistic Stealth2025.0552.92131.76-1.1890100
StegoAttackTarget Model=DeepSeek-V3.2, Stealth Category=Fully Stealth2025.0553.41981.560.51221
FFATarget Model=DeepSeek-V3.2, Stealth Category=Zero Stealth2025.05115.35734.04-0.919267
GCGTarget Model=DeepSeek-V3.2, Stealth Category=Zero Stealth2025.052,877.84055.66-1.2110033