Jailbreak Attack on HarmBench target: Phi-4-Mini
73ASRAGILE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AGILEGenerator LLM=Llama-3-8B-Instruct, p=5, NCand=25, Evaluation Judge=GPT-Judge2025.08 | 73 | 4.6 | |
| AutoDANEvaluation Judge=GPT-Judge2025.08 | 53 | 4.31 | |
| ActorBreakerEvaluation Judge=GPT-Judge2025.08 | 24 | 3.39 | |
| CrescendoEvaluation Judge=GPT-Judge2025.08 | 22 | 3.15 | |
| ReNeLLMEvaluation Judge=GPT-Judge2025.08 | 21.5 | 3.3 | |
| PAIREvaluation Judge=GPT-Judge2025.08 | 18 | 3.29 | |
| GCGEvaluation Judge=GPT-Judge2025.08 | 16.5 | 2.37 | |
| PiFEvaluation Judge=GPT-Judge2025.08 | 6 | 2.19 | |
| I-GCGEvaluation Judge=GPT-Judge2025.08 | 5.5 | 1.58 | |
| AmpleGCGEvaluation Judge=GPT-Judge2025.08 | 5 | 1.63 | |
| CoAEvaluation Judge=GPT-Judge2025.08 | 3 | 1.87 |