Jailbreak Attack Stealth Evaluation on AdvBench 50
4.4039PPLArtPrompt
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| ArtPromptTarget Model=DeepSeek-V3.2, Stealth Category=Semantic Stealth2025.05 | 4.4039 | 48.5 | -0.35 | 85 | 100 | |
| JailbrokenTarget Model=DeepSeek-V3.2, Stealth Category=Zero Stealth2025.05 | 5.5776 | 2.04 | -1.16 | 90 | 100 | |
| DRATarget Model=DeepSeek-V3.2, Stealth Category=Semantic Stealth2025.05 | 14.6255 | 78.86 | -0.68 | 100 | 88 | |
| DrAttackTarget Model=DeepSeek-V3.2, Stealth Category=Semantic Stealth2025.05 | 17.9678 | 11.26 | -0.04 | 83 | 89 | |
| CipherTarget Model=DeepSeek-V3.2, Stealth Category=Semantic Stealth2025.05 | 32.5532 | 160.8 | 0.47 | 0 | 17 | |
| VanillaTarget Model=DeepSeek-V3.22025.05 | 38.9457 | 0.04 | -1.6 | 100 | 75 | |
| AutoDANTarget Model=DeepSeek-V3.2, Stealth Category=Linguistic Stealth2025.05 | 52.9213 | 1.76 | -1.18 | 90 | 100 | |
| StegoAttackTarget Model=DeepSeek-V3.2, Stealth Category=Fully Stealth2025.05 | 53.4198 | 1.56 | 0.51 | 2 | 21 | |
| FFATarget Model=DeepSeek-V3.2, Stealth Category=Zero Stealth2025.05 | 115.3573 | 4.04 | -0.91 | 92 | 67 | |
| GCGTarget Model=DeepSeek-V3.2, Stealth Category=Zero Stealth2025.05 | 2,877.8405 | 5.66 | -1.21 | 100 | 33 |