Jailbreak Attack on FinRisk-Bench
98.08Llama 3 70BCoRT (RCA+RCC)
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| CoRT (RCA+RCC)Attack Method Type=Multi-Turn, Maximum Rounds=52025.09 | 98.08 | 98.66 | 97.32 | 98.66 | 98.66 | 89.46 | 98.28 | 78.54 | 95 | |
| CoRT (RCA)Attack Method Type=Multi-Turn, Maximum Rounds=52025.09 | 97.7 | 98.47 | 95.21 | 97.51 | 98.28 | 85.06 | 97.89 | 71.07 | 93.19 | |
| X-TeamingAttack Method Type=Multi-Turn, Maximum Rounds=52025.09 | 88.31 | 93.68 | 86.97 | 91.38 | 92.53 | 84.48 | 85.06 | 81.61 | 88.36 | |
| ReNeLLMAttack Method Type=Single-Turn2025.09 | 85.05 | 77.2 | 94.06 | 89.85 | 93.68 | 70.5 | 31.42 | 6.89 | 71.59 | |
| DeepInceptionAttack Method Type=Single-Turn2025.09 | 69.92 | 36.02 | 58.62 | 42.34 | 61.11 | 9.39 | 4.98 | 0 | 34.01 | |
| CodeChameleonAttack Method Type=Single-Turn2025.09 | 67.24 | 65.71 | 84.1 | 74.05 | 93.1 | 89.27 | 6.7 | 77.78 | 71.88 | |
| FITDAttack Method Type=Multi-Turn, Maximum Rounds=52025.09 | 61.88 | 71.84 | 65.9 | 77.01 | 63.79 | 75.29 | 79.5 | 68.97 | 72.06 | |
| CrescendoAttack Method Type=Multi-Turn, Maximum Rounds=52025.09 | 60.15 | 36.59 | 29.69 | 27.78 | 28.35 | 5.94 | 7.85 | 0 | 26.84 | |
| FlipAttackAttack Method Type=Single-Turn2025.09 | 42.72 | 93.68 | 92.53 | 82.57 | 94.83 | 81.61 | 0.57 | 0 | 64.62 | |
| ArtPromptAttack Method Type=Single-Turn2025.09 | 40.8 | 39.08 | 13.03 | 36.21 | 43.49 | 8.62 | 0 | 0.19 | 26.71 | |
| ActorAttackAttack Method Type=Multi-Turn, Maximum Rounds=52025.09 | 35.44 | 60.54 | 70.88 | 61.3 | 70.88 | 54.79 | 64.56 | 43.49 | 59.13 |