Jailbreak Attack on HarmBench target: Llama-3.1-8B
75Attack Success Rate (ASR)AutoDAN
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AutoDANEvaluation Judge=GPT-Judge2025.08 | 75 | 4.57 | |
| AGILEGenerator LLM=Llama-3-8B-Instruct, p=5, NCand=25, Evaluation Judge=GPT-Judge2025.08 | 68.5 | 4.58 | |
| ReNeLLMEvaluation Judge=GPT-Judge2025.08 | 59 | 4.26 | |
| ActorBreakerEvaluation Judge=GPT-Judge2025.08 | 46 | 3.6 | |
| CrescendoEvaluation Judge=GPT-Judge2025.08 | 33.5 | 3.58 | |
| PAIREvaluation Judge=GPT-Judge2025.08 | 18 | 3.21 | |
| GCGEvaluation Judge=GPT-Judge2025.08 | 16.5 | 1.44 | |
| PiFEvaluation Judge=GPT-Judge2025.08 | 5.5 | 1.37 | |
| AmpleGCGEvaluation Judge=GPT-Judge2025.08 | 4 | 1.32 | |
| CoAEvaluation Judge=GPT-Judge2025.08 | 2 | 1.63 | |
| I-GCGEvaluation Judge=GPT-Judge2025.08 | 1.5 | 1.17 |