Jailbreak Attack on JailbreakBench (ASR, BERTScore)
91ASRSAHA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SAHATarget Model=Deepseek-llm-7B-Chat2026.03 | 91 | 0.7 | |
| SAHATarget Model=Llama3.1-8B-Instruct2026.03 | 85 | 0.76 | |
| SCAVTarget Model=Deepseek-llm-7B-Chat2026.03 | 85 | 0.71 | |
| SAHATarget Model=Qwen1.5-7B-Chat2026.03 | 82 | 0.75 | |
| SCAVTarget Model=Qwen1.5-7B-Chat2026.03 | 71 | 0.73 | |
| AutoDAN-TurboTarget Model=Qwen1.5-7B-Chat2026.03 | 63 | 0.75 | |
| AutoDANTarget Model=Deepseek-llm-7B-Chat2026.03 | 60 | 0.55 | |
| ConVATarget Model=Qwen1.5-7B-Chat2026.03 | 59 | 0.4 | |
| PAIRTarget Model=Llama3.1-8B-Instruct2026.03 | 57 | 0.07 | |
| AutoDANTarget Model=Qwen1.5-7B-Chat2026.03 | 57 | 0.62 | |
| SCAVTarget Model=Llama3.1-8B-Instruct2026.03 | 55 | 0.77 | |
| GCGTarget Model=Qwen1.5-7B-Chat2026.03 | 47 | 0.57 | |
| AutoDANTarget Model=Llama3.1-8B-Instruct2026.03 | 40 | 0.58 | |
| CAATarget Model=Qwen1.5-7B-Chat2026.03 | 39 | 0.38 | |
| GCGTarget Model=Deepseek-llm-7B-Chat2026.03 | 37 | 0.64 | |
| CAATarget Model=Deepseek-llm-7B-Chat2026.03 | 37 | 0.15 | |
| PAIRTarget Model=Deepseek-llm-7B-Chat2026.03 | 30 | 0.59 | |
| BasedTarget Model=Deepseek-llm-7B-Chat2026.03 | 29 | 0.64 | |
| ConVATarget Model=Llama3.1-8B-Instruct2026.03 | 19 | 0.39 | |
| CAATarget Model=Llama3.1-8B-Instruct2026.03 | 14 | 0.35 | |
| AutoDAN-TurboTarget Model=Llama3.1-8B-Instruct2026.03 | 13 | 0.7 | |
| AutoDAN-TurboTarget Model=Deepseek-llm-7B-Chat2026.03 | 13 | 0.11 | |
| GCGTarget Model=Llama3.1-8B-Instruct2026.03 | 10 | 0.29 | |
| PAIRTarget Model=Qwen1.5-7B-Chat2026.03 | 10 | 0.23 | |
| BasedTarget Model=Llama3.1-8B-Instruct2026.03 | 6 | 0.62 | |
| BasedTarget Model=Qwen1.5-7B-Chat2026.03 | 6 | 0.74 | |
| ConVATarget Model=Deepseek-llm-7B-Chat2026.03 | 3 | 0.4 |