Jailbreak Detection on PAIR
98AccuracyIMAG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| IMAGTarget LLM=Vicuna-13B2025.12 | 98 | 99 | |
| IMAGTarget LLM=Mistral-7B2025.12 | 93 | 96 | |
| IMAGTarget LLM=Vicuna-7B2025.12 | 88 | 93 | |
| IMAGTarget LLM=Llama3-8B2025.12 | 78 | 87 | |
| IMAGTarget LLM=Llama2-7B2025.12 | 76 | 87 | |
| LlamaGTarget LLM=Vicuna-7B2025.12 | 75 | 85 | |
| LlamaGTarget LLM=Vicuna-13B2025.12 | 75 | 85 | |
| LlamaGTarget LLM=Mistral-7B2025.12 | 74 | 85 | |
| LlamaGTarget LLM=Llama3-8B2025.12 | 70 | 82 | |
| GradSafeTarget LLM=Llama2-7B2025.12 | 62 | 77 | |
| LlamaGTarget LLM=Llama2-7B2025.12 | 53 | 69 | |
| Self-ExTarget LLM=Mistral-7B2025.12 | 46 | 63 | |
| Self-ExTarget LLM=Llama2-7B2025.12 | 32 | 35 | |
| Self-ExTarget LLM=Llama3-8B2025.12 | 16 | 27 | |
| PPLTarget LLM=Llama2-7B2025.12 | 10 | 18 | |
| OAPITarget LLM=Mistral-7B2025.12 | 7 | 13 | |
| OAPITarget LLM=Vicuna-13B2025.12 | 5 | 10 | |
| OAPITarget LLM=Vicuna-7B2025.12 | 4 | 9 | |
| Self-ExTarget LLM=Vicuna-7B2025.12 | 3 | 6 | |
| GradSafeTarget LLM=Vicuna-7B2025.12 | 3 | 6 | |
| OAPITarget LLM=Llama2-7B2025.12 | 3 | 5 | |
| OAPITarget LLM=Llama3-8B2025.12 | 3 | 6 | |
| PPLTarget LLM=Vicuna-13B2025.12 | 1 | 2 | |
| PPLTarget LLM=Mistral-7B2025.12 | 0 | 0 | |
| GradSafeTarget LLM=Mistral-7B2025.12 | 0 | 0 | |
| PPLTarget LLM=Vicuna-7B2025.12 | 0 | 0 | |
| Self-ExTarget LLM=Vicuna-13B2025.12 | 0 | 0 | |
| GradSafeTarget LLM=Vicuna-13B2025.12 | 0 | 0 | |
| PPLTarget LLM=Llama3-8B2025.12 | 0 | 0 | |
| GradSafeTarget LLM=Llama3-8B2025.12 | 0 | 0 |