Pairwise Prediction Success Rate on Jailbreak Prompt Dataset
78.95Proxy ASRNarrow Safety Proxy
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Narrow Safety ProxyTarget Model=Llama2025.11 | 78.95 | 90.71 | |
| Narrow Safety ProxyTarget Model=Qwen2025.11 | 73.49 | 79.23 | |
| Narrow Safety ProxyTarget Model=GPT-4o Mini2025.11 | 69.16 | 91.1 |