Jailbreak Attack on HarmBench (Qwen2.5-7B Target)
3ASRI-GCG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| I-GCGEvaluation Judge=GPT-Judge2025.08 | 3 | 1.45 | |
| CoAEvaluation Judge=GPT-Judge2025.08 | 6 | 1.92 | |
| GCGEvaluation Judge=GPT-Judge2025.08 | 6.5 | 1.82 | |
| AmpleGCGEvaluation Judge=GPT-Judge2025.08 | 8 | 1.76 | |
| PiFEvaluation Judge=GPT-Judge2025.08 | 9.5 | 2.42 | |
| PAIREvaluation Judge=GPT-Judge2025.08 | 29.5 | 3.61 | |
| CrescendoEvaluation Judge=GPT-Judge2025.08 | 35.5 | 3.46 | |
| ActorBreakerEvaluation Judge=GPT-Judge2025.08 | 47 | 3.89 | |
| ReNeLLMEvaluation Judge=GPT-Judge2025.08 | 53.5 | 4.23 | |
| AutoDANEvaluation Judge=GPT-Judge2025.08 | 77 | 4.67 | |
| AGILEGenerator LLM=Llama-3-8B-Instruct, p=5, NCand=25, Evaluation Judge=GPT-Judge2025.08 | 89 | 4.85 |