Jailbreak Evaluation on AdvBench Ensemble configuration Claude-v2
1.08Harmfulness Score (HS)ArtPrompt
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ArtPromptDefense=self-reminder2024.12 | 1.08 | 0 | |
| SATA-ELPDefense=self-reminder2024.12 | 1.44 | 6 | |
| ArtPromptDefense=paraphrase2024.12 | 1.6 | 8 | |
| SATA-MLMDefense=self-reminder2024.12 | 1.66 | 8 | |
| ArtPromptDefense=rpo2024.12 | 2.22 | 20 | |
| SATA-MLMDefense=paraphrase2024.12 | 3.36 | 38 | |
| SATA-MLMDefense=rpo2024.12 | 3.38 | 56 | |
| ArtPromptDefense=None2024.12 | 3.44 | 52 | |
| ArtPromptDefense=windowed PPL-filter2024.12 | 3.64 | 40 | |
| SATA-MLMDefense=windowed PPL-filter2024.12 | 3.8 | 66 | |
| SATA-MLMDefense=None2024.12 | 3.86 | 68 | |
| SATA-ELPDefense=paraphrase2024.12 | 4.06 | 64 | |
| SATA-ELPDefense=rpo2024.12 | 4.28 | 72 | |
| SATA-ELPDefense=windowed PPL-filter2024.12 | 4.48 | 84 | |
| SATA-ELPDefense=None2024.12 | 4.54 | 86 |