Jailbreak Defense on Aggregate Benchmarks
1.06Harmful ScoreSAGE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SAGEModel=Llama3.12025.05 | 1.06 | 1 | |
| SAGEModel=Qwen2.52025.05 | 1.08 | 1 | |
| Goal PrioritizationModel=Qwen2.52025.05 | 1.17 | 5 | |
| Self-ExaminationModel=Llama3.12025.05 | 1.18 | 7 | |
| SAGEModel=Gemma22025.05 | 1.28 | 5 | |
| Goal PrioritizationModel=Llama3.12025.05 | 1.33 | 22 | |
| Goal PrioritizationModel=Gemma22025.05 | 1.45 | 16 | |
| Self-ExaminationModel=Gemma22025.05 | 1.67 | 25 | |
| Self-ExaminationModel=Qwen2.52025.05 | 1.68 | 12 | |
| ICDModel=Llama3.12025.05 | 1.71 | 11 | |
| Self-ReminderModel=Llama3.12025.05 | 1.77 | 25 | |
| IAModel=Llama3.12025.05 | 1.89 | 23 | |
| Self-ReminderModel=Qwen2.52025.05 | 2.24 | 43 | |
| IAModel=Qwen2.52025.05 | 2.27 | 41 | |
| ICDModel=Qwen2.52025.05 | 2.53 | 50 | |
| IAModel=Gemma22025.05 | 2.63 | 49 | |
| Self-ReminderModel=Gemma22025.05 | 3.06 | 50 | |
| No DefenseModel=Llama3.12025.05 | 3.2 | 57 | |
| ICDModel=Gemma22025.05 | 3.67 | 66 | |
| No DefenseModel=Qwen2.52025.05 | 3.68 | 67 | |
| No DefenseModel=Gemma22025.05 | 4.01 | 74 |