Jailbreak on JBB-Behaviors utilitarian dilemmas (test)
87Jailbreak Success RateTRIAL
Evaluation Results
| Method | Links | |
|---|---|---|
| TRIALTurns=Multi, Target Model=GLM-4-Plus2025.09 | 87 | |
| DRATurns=Single, Target Model=GLM-4-Plus2025.09 | 83 | |
| JigsawTurns=Multi, Target Model=GLM-4-Plus2025.09 | 79 | |
| PAIRTurns=Single, Target Model=Vicuna-13B2025.09 | 76 | |
| TRIALTurns=Multi, Target Model=DeepSeek-V32025.09 | 76 | |
| TRIALTurns=Multi, Target Model=Llama-3.1-8B2025.09 | 70 | |
| FITDTurns=Multi, Target Model=DeepSeek-V32025.09 | 69 | |
| JigsawTurns=Multi, Target Model=DeepSeek-V32025.09 | 67 | |
| PAIRTurns=Single, Target Model=GLM-4-Plus2025.09 | 64 | |
| JigsawTurns=Multi, Target Model=Vicuna-13B2025.09 | 64 | |
| PAIRTurns=Single, Target Model=GPT-3.52025.09 | 63 | |
| DRATurns=Single, Target Model=GPT-3.52025.09 | 57 | |
| ActorAttackTurns=Multi, Target Model=Vicuna-13B2025.09 | 54 | |
| TRIALTurns=Multi, Target Model=Vicuna-13B2025.09 | 52 | |
| DRATurns=Single, Target Model=DeepSeek-V32025.09 | 51 | |
| PAIRTurns=Single, Target Model=DeepSeek-V32025.09 | 50 | |
| ActorAttackTurns=Multi, Target Model=DeepSeek-V32025.09 | 50 | |
| ActorAttackTurns=Multi, Target Model=GPT-3.52025.09 | 49 | |
| JigsawTurns=Multi, Target Model=Llama-3.1-8B2025.09 | 49 | |
| ActorAttackTurns=Multi, Target Model=GLM-4-Plus2025.09 | 47 | |
| TRIALTurns=Multi, Target Model=GPT-4o2025.09 | 46 | |
| JigsawTurns=Multi, Target Model=GPT-3.52025.09 | 43 | |
| DRATurns=Single, Target Model=Llama-3.1-8B2025.09 | 42 | |
| JigsawTurns=Multi, Target Model=GPT-42025.09 | 42 | |
| PAIRTurns=Single, Target Model=GPT-4o2025.09 | 39 | |
| ActorAttackTurns=Multi, Target Model=GPT-4o2025.09 | 39 | |
| FITDTurns=Multi, Target Model=GPT-4o2025.09 | 38 | |
| ActorAttackTurns=Multi, Target Model=Claude-3.72025.09 | 37 | |
| DRATurns=Single, Target Model=GPT-42025.09 | 35 | |
| GCGTurns=Single, Target Model=GPT-3.52025.09 | 34 | |
| TRIALTurns=Multi, Target Model=GPT-42025.09 | 34 | |
| GCGTurns=Single, Target Model=Vicuna-13B2025.09 | 30 | |
| ActorAttackTurns=Multi, Target Model=GPT-42025.09 | 30 | |
| DeepInceptionTurns=Single, Target Model=GLM-4-Plus2025.09 | 29 | |
| FITDTurns=Multi, Target Model=GLM-4-Plus2025.09 | 29 | |
| FITDTurns=Multi, Target Model=Claude-3.72025.09 | 29 | |
| ActorAttackTurns=Multi, Target Model=Llama-3.1-8B2025.09 | 28 | |
| FITDTurns=Multi, Target Model=Vicuna-13B2025.09 | 27 | |
| FITDTurns=Multi, Target Model=GPT-42025.09 | 26 | |
| TRIALTurns=Multi, Target Model=GPT-3.52025.09 | 23 | |
| PAIRTurns=Single, Target Model=Llama-3.1-8B2025.09 | 19 | |
| PAPTurns=Single, Target Model=GPT-3.52025.09 | 16 | |
| PAIRTurns=Single, Target Model=GPT-42025.09 | 16 | |
| DeepInceptionTurns=Single, Target Model=DeepSeek-V32025.09 | 16 | |
| TRIALTurns=Multi, Target Model=Claude-3.72025.09 | 14 | |
| FITDTurns=Multi, Target Model=Llama-3.1-8B2025.09 | 12 | |
| FITDTurns=Multi, Target Model=GPT-3.52025.09 | 12 | |
| GCGTurns=Single, Target Model=Llama-3.1-8B2025.09 | 11 | |
| GCGTurns=Single, Target Model=DeepSeek-V32025.09 | 8 | |
| GCGTurns=Single, Target Model=GLM-4-Plus2025.09 | 6 | |
| PAPTurns=Single, Target Model=Vicuna-13B2025.09 | 6 | |
| DRATurns=Single, Target Model=GPT-4o2025.09 | 5 | |
| DeepInceptionTurns=Single, Target Model=GPT-3.52025.09 | 5 | |
| JigsawTurns=Multi, Target Model=GPT-4o2025.09 | 5 | |
| PAPTurns=Single, Target Model=Llama-3.1-8B2025.09 | 3 | |
| PAPTurns=Single, Target Model=Claude-3.72025.09 | 3 | |
| DeepInceptionTurns=Single, Target Model=Vicuna-13B2025.09 | 3 | |
| DeepInceptionTurns=Single, Target Model=GPT-42025.09 | 3 | |
| PAIRTurns=Single, Target Model=Claude-3.72025.09 | 2 | |
| PAPTurns=Single, Target Model=GPT-4o2025.09 | 1 | |
| DeepInceptionTurns=Single, Target Model=Llama-3.1-8B2025.09 | 1 | |
| DeepInceptionTurns=Single, Target Model=GPT-4o2025.09 | 1 | |
| JigsawTurns=Multi, Target Model=Claude-3.72025.09 | 1 | |
| GCGTurns=Single, Target Model=GPT-42025.09 | 0 | |
| GCGTurns=Single, Target Model=GPT-4o2025.09 | 0 | |
| GCGTurns=Single, Target Model=Claude-3.72025.09 | 0 | |
| PAPTurns=Single, Target Model=DeepSeek-V32025.09 | 0 | |
| PAPTurns=Single, Target Model=GPT-42025.09 | 0 | |
| PAPTurns=Single, Target Model=GLM-4-Plus2025.09 | 0 | |
| DRATurns=Single, Target Model=Vicuna-13B2025.09 | 0 | |
| DRATurns=Single, Target Model=Claude-3.72025.09 | 0 | |
| DeepInceptionTurns=Single, Target Model=Claude-3.72025.09 | 0 |