Jailbreak Attack on HarmBench target: Llama-3.2-3B
72.5ASRAGILE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AGILEGenerator LLM=Llama-3-8B-Instruct, p=5, NCand=25, Evaluation Judge=GPT-Judge2025.08 | 72.5 | 4.66 | |
| AutoDANEvaluation Judge=GPT-Judge2025.08 | 58.5 | 4.37 | |
| ActorBreakerEvaluation Judge=GPT-Judge2025.08 | 33 | 3.01 | |
| ReNeLLMEvaluation Judge=GPT-Judge2025.08 | 31.5 | 3.41 | |
| CrescendoEvaluation Judge=GPT-Judge2025.08 | 25 | 3.32 | |
| PiFEvaluation Judge=GPT-Judge2025.08 | 18 | 2.24 | |
| PAIREvaluation Judge=GPT-Judge2025.08 | 13.5 | 2.82 | |
| CoAEvaluation Judge=GPT-Judge2025.08 | 4 | 1.2 | |
| GCGEvaluation Judge=GPT-Judge2025.08 | 3 | 1.51 | |
| AmpleGCGEvaluation Judge=GPT-Judge2025.08 | 3 | 1.28 | |
| I-GCGEvaluation Judge=GPT-Judge2025.08 | 1.5 | 1.32 |