Jailbreak Attack on JailbreakBench PAIR
82Attack Success Rate (ASR)None
Evaluation Results
| Method | Links | |
|---|---|---|
| NoneModel=Vicuna2024.01 | 82 | |
| Perplexity FilterModel=Vicuna2024.01 | 81 | |
| SmoothLLMModel=Vicuna2024.01 | 47 | |
| w/o DefenseTarget Model=GPT-4o2026.04 | 30 | |
| RephrasingModel=Vicuna2024.01 | 25 | |
| RPOModel=Vicuna2024.01 | 20 | |
| Smooth LLMTarget Model=GPT-4o2026.04 | 19 | |
| Avg. ASRTarget Model=GPT-4o2026.04 | 16 | |
| Self ReminderTarget Model=GPT-4o2026.04 | 11 | |
| NoneModel=Llama-22024.01 | 4 | |
| Perplexity FilterModel=Llama-22024.01 | 4 | |
| RephrasingModel=Llama-22024.01 | 4 | |
| SmoothLLMModel=Llama-22024.01 | 1 | |
| Input FilterTarget Model=GPT-4o2026.04 | 1 | |
| RPOModel=Llama-22024.01 | 0 |