Jailbreaking on HarmBench n = 200
16ASRGCG
Evaluation Results
| Method | Links | |
|---|---|---|
| GCGTarget Model=Qwen3-8B, Iterations=@82026.07 | 16 | |
| GCGTarget Model=GLM-4-9B, Iterations=@82026.07 | 19 | |
| GCGTarget Model=Llama-3-8B, Iterations=@82026.07 | 19.5 | |
| CSRTTarget Model=DeepSeek-R1, Iterations=@82026.07 | 32.5 | |
| GCGTarget Model=DeepSeek-R1, Iterations=@82026.07 | 41.5 | |
| CSRTTarget Model=Qwen3-8B, Iterations=@82026.07 | 49.5 | |
| CSRTTarget Model=Llama-3-8B, Iterations=@82026.07 | 60.5 | |
| CSRTTarget Model=Gemma-7B, Iterations=@82026.07 | 66 | |
| STEERTarget Model=DeepSeek-R1, Iterations=@82026.07 | 66 | |
| GCGTarget Model=Gemma-7B, Iterations=@82026.07 | 73 | |
| STEERTarget Model=Qwen3-8B, Iterations=@82026.07 | 76 | |
| STEERTarget Model=Llama-3-8B, Iterations=@82026.07 | 76.5 | |
| CSRTTarget Model=Mistral-7B, Iterations=@82026.07 | 78.5 | |
| CSRTTarget Model=GLM-4-9B, Iterations=@82026.07 | 81.5 | |
| GCGTarget Model=Mistral-7B, Iterations=@82026.07 | 82.5 | |
| STEERTarget Model=Gemma-7B, Iterations=@82026.07 | 86 | |
| STEERTarget Model=GLM-4-9B, Iterations=@82026.07 | 86 | |
| STEERTarget Model=Mistral-7B, Iterations=@82026.07 | 87.5 |