Jailbreak on AdvBench Ensemble configuration GPT-4o
0Attack Success Rate (ASR)ArtPrompt
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ArtPromptDefense=self-reminder2024.12 | 0 | 1.06 | |
| SATA-ELPDefense=self-reminder2024.12 | 2 | 1.38 | |
| CipherChatlow_perplexity=false, single_turn_attack=true, require_human_strategies=true2025.12 | 10 | — | |
| GCGlow_perplexity=false, single_turn_attack=true, require_human_strategies=false2025.12 | 12.5 | — | |
| SATA-MLMDefense=self-reminder2024.12 | 26 | 2.28 | |
| ArtPromptDefense=windowed PPL-filter2024.12 | 30 | 2.88 | |
| PAIRlow_perplexity=true, single_turn_attack=true, require_human_strategies=true2025.12 | 39 | — | |
| ArtPromptDefense=paraphrase2024.12 | 42 | 3.88 | |
| PAPlow_perplexity=true, single_turn_attack=true, require_human_strategies=true2025.12 | 42 | — | |
| ArtPromptDefense=None2024.12 | 48 | 3.38 | |
| ArtPromptDefense=rpo2024.12 | 52 | 3.88 | |
| SATA-ELPDefense=rpo2024.12 | 68 | 4.28 | |
| SATA-MLMDefense=paraphrase2024.12 | 70 | 4.22 | |
| Code Attacklow_perplexity=true, single_turn_attack=true, require_human_strategies=true2025.12 | 70.5 | — | |
| Bijectionlow_perplexity=false, single_turn_attack=true, require_human_strategies=false2025.12 | 72.3 | — | |
| SATA-MLMDefense=rpo2024.12 | 74 | 4.2 | |
| SATA-ELPDefense=None2024.12 | 78 | 4.56 | |
| SATA-ELPDefense=windowed PPL-filter2024.12 | 78 | 4.48 | |
| SATA-ELPDefense=paraphrase2024.12 | 78 | 4.62 | |
| SATA-MLMDefense=None2024.12 | 82 | 4.36 | |
| SATA-MLMDefense=windowed PPL-filter2024.12 | 84 | 4.54 | |
| ActorAttacklow_perplexity=true, single_turn_attack=false, require_human_strategies=false2025.12 | 84.5 | — | |
| BoN Attacklow_perplexity=false, single_turn_attack=true, require_human_strategies=false2025.12 | 88.7 | — | |
| J2 Attacklow_perplexity=true, single_turn_attack=false, require_human_strategies=true2025.12 | 97.5 | — | |
| Jailbreak-Zerolow_perplexity=true, single_turn_attack=true, require_human_strategies=false2025.12 | 99.5 | — |