Jailbreak Defense on ReNeLLM, DeepInception, GPTFuzzer, CodeAttack (test)
1Harmful Score (ReNeLLM)SAGE
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| SAGEModel=deepseek R12025.05 | 1 | 0 | 1 | 0 | 1 | 0 | 1 | 0 | 1 | 0 | |
| Goal PrioritizationModel=deepseek R12025.05 | 1.04 | 2 | 1 | 40 | 1.16 | 12 | 1 | 0 | 1.05 | 14 | |
| IAModel=deepseek R12025.05 | 1.5 | 22 | 1 | 18 | 1.4 | 10 | 1.08 | 18 | 1.25 | 17 | |
| Self-ReminderModel=deepseek R12025.05 | 2.08 | 86 | 1.34 | 98 | 2.14 | 32 | 1 | 56 | 1.64 | 68 | |
| ICDModel=deepseek R12025.05 | 3.02 | 56 | 2.82 | 80 | 3.02 | 30 | 1.34 | 20 | 2.55 | 46 | |
| No DefenseModel=deepseek R12025.05 | 4.68 | 96 | 4.88 | 98 | 3.66 | 42 | 4.48 | 100 | 4.42 | 84 |