Jail-breaking detection on WildChat
0.895AUC (Statistical Power)RUT
Evaluation Results
| Method | Links | |
|---|---|---|
| RUTModel=Mistral, Prompt=Dan2025.06 | 0.895 | |
| RUTModel=Mistral, Prompt=Anti-Dan2025.06 | 0.893 | |
| RUTModel=Gemma, Prompt=Evil-Bot2025.06 | 0.893 | |
| RUTModel=Mistral, Prompt=Evil-Bot2025.06 | 0.892 | |
| RUTModel=Gemma, Prompt=Dan2025.06 | 0.888 | |
| KSModel=Mistral, Prompt=Dan2025.06 | 0.873 | |
| KSModel=Mistral, Prompt=Evil-Bot2025.06 | 0.873 | |
| KSModel=Mistral, Prompt=Anti-Dan2025.06 | 0.872 | |
| KSModel=Gemma, Prompt=Evil-Bot2025.06 | 0.871 | |
| KSModel=Gemma, Prompt=Dan2025.06 | 0.867 | |
| RUTModel=Gemma, Prompt=Anti-Dan2025.06 | 0.858 | |
| KSModel=Gemma, Prompt=Anti-Dan2025.06 | 0.854 | |
| MMDModel=Gemma, Prompt=Anti-Dan2025.06 | 0.816 | |
| MMDModel=Mistral, Prompt=Dan2025.06 | 0.802 | |
| MMDModel=Mistral, Prompt=Anti-Dan2025.06 | 0.781 | |
| MMDModel=Mistral, Prompt=Evil-Bot2025.06 | 0.766 | |
| MMDModel=Gemma, Prompt=Dan2025.06 | 0.757 | |
| MMDModel=Gemma, Prompt=Evil-Bot2025.06 | 0.753 |