Jailbreak Attack on HarmBench Vicuna (test)
96Attack Success Rate (ASR)Random
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RandomAttack=GCG2025.02 | 96 | 39 | 82.13 | 0.8 | |
| BaselineAttack=GCG2025.02 | 96 | 12 | 37.4 | 0.68 | |
| StandardAttack=GCG2025.02 | 98 | 30 | 85.29 | 0.86 | |
| 25-CRIAttack=GCG2025.02 | 98 | 3 | 26.71 | 0.58 | |
| 1-CRIAttack=GCG2025.02 | 98 | 1 | 10.51 | 0.54 | |
| 5-CRIAttack=GCG2025.02 | 100 | 6 | 19 | 0.6 | |
| StandardAttack=Embedding2025.02 | 100 | 13 | 14.36 | 1.78 | |
| RandomAttack=Embedding2025.02 | 100 | 15 | 15.2 | 1.73 | |
| BaselineAttack=Embedding2025.02 | 100 | 3 | 3.96 | 0.95 | |
| 25-CRIAttack=Embedding2025.02 | 100 | 2 | 2.84 | 0.35 | |
| 5-CRIAttack=Embedding2025.02 | 100 | 2 | 2.58 | 0.51 | |
| 1-CRIAttack=Embedding2025.02 | 100 | 2.5 | 2.68 | 0.61 |