Jailbreak Defense and Utility Evaluation on Implicit Risk Finance Dataset
0.4Jailbreak Success Rate (JSR)MENTOR
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MENTORModel=Qwen3-235B2025.11 | 0.4 | 4.918 | |
| DPP-MENTORModel=GPT-4o2025.11 | 0.6 | 4.368 | |
| MENTORModel=DeepSeek-R12025.11 | 0.6 | 4.856 | |
| DPP-MENTORModel=Qwen3-235B2025.11 | 1 | 4.908 | |
| MENTORModel=GPT-4o2025.11 | 1.6 | 4.386 | |
| Self-ReminderModel=Qwen3-235B2025.11 | 1.6 | 4.544 | |
| DPP-MENTORModel=DeepSeek-R12025.11 | 2.4 | 4.776 | |
| G4D-RuleTreeModel=Qwen3-235B2025.11 | 2.8 | 4.516 | |
| G4DModel=Qwen3-235B2025.11 | 3 | 4.642 | |
| G4D-RuleTreeModel=DeepSeek-R12025.11 | 5.8 | 4.7 | |
| G4D-RuleTreeModel=GPT-4o2025.11 | 6.8 | 4.472 | |
| Self-ReminderModel=GPT-4o2025.11 | 8 | 4.396 | |
| G4DModel=GPT-4o2025.11 | 9.2 | 4.534 | |
| G4DModel=DeepSeek-R12025.11 | 9.4 | 4.72 | |
| Self-ReminderModel=DeepSeek-R12025.11 | 12 | 4.678 | |
| DPPModel=Qwen3-235B2025.11 | 26.8 | 4.77 | |
| ProBModel=Qwen3-235B2025.11 | 28.8 | 4.316 | |
| RawModel=Qwen3-235B2025.11 | 30 | 4.854 | |
| RPO-MENTORModel=Qwen3-235B2025.11 | 32.2 | 4.796 | |
| RPOModel=Qwen3-235B2025.11 | 39.2 | 4.892 | |
| ProBModel=DeepSeek-R12025.11 | 47.8 | 4.456 | |
| RPO-MENTORModel=DeepSeek-R12025.11 | 48.4 | 4.78 | |
| RawModel=DeepSeek-R12025.11 | 52.2 | 4.924 | |
| RPOModel=DeepSeek-R12025.11 | 57.2 | 4.878 | |
| DPPModel=DeepSeek-R12025.11 | 58.6 | 4.922 | |
| RPO-MENTORModel=GPT-4o2025.11 | 63.2 | 4.166 | |
| ProBModel=GPT-4o2025.11 | 64.2 | 4.066 | |
| RPOModel=GPT-4o2025.11 | 72.6 | 4.442 | |
| DPPModel=GPT-4o2025.11 | 75 | 4.564 | |
| RawModel=GPT-4o2025.11 | 82.6 | 4.65 |