Adversarial Attack on StrongREJECT Hijacked (test)
0CHRFPFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FPFTModel=DeepSeek-7B2026.04 | 0 | 25 | |
| OriginalModel=Llama3.2-3B2026.04 | 1 | 4 | |
| AutoRANModel=Qwen2.5-7B2026.04 | 4 | 15 | |
| AutoRANModel=Llama3.2-3B2026.04 | 16 | 29 | |
| AutoRANModel=DeepSeek-7B2026.04 | 17 | 22 | |
| CoT HijackingModel=Llama3.2-3B2026.04 | 22 | 9 | |
| OriginalModel=DeepSeek-7B2026.04 | 24 | 30 | |
| CoT HijackingModel=DeepSeek-7B2026.04 | 28 | 22 | |
| CoT HijackingModel=Qwen2.5-7B2026.04 | 31 | 24 | |
| Chain-of-LureModel=Llama3.2-3B2026.04 | 33 | 30 | |
| Chain-of-LureModel=DeepSeek-7B2026.04 | 42 | 41 | |
| H-CoTModel=DeepSeek-7B2026.04 | 45 | 32 | |
| H-CoTModel=Qwen2.5-7B2026.04 | 46 | 26 | |
| H-CoTModel=Llama3.2-3B2026.04 | 46 | 7 | |
| OriginalModel=Qwen2.5-7B2026.04 | 49 | 9 | |
| Chain-of-LureModel=Qwen2.5-7B2026.04 | 51 | 41 | |
| LRFTModel=Qwen2.5-7B2026.04 | 62 | 62 | |
| FPFTModel=Llama3.2-3B2026.04 | 80 | 79 | |
| LRFTModel=Llama3.2-3B2026.04 | 81 | 98 | |
| FPFTModel=Qwen2.5-7B2026.04 | 82 | 89 | |
| LRFTModel=DeepSeek-7B2026.04 | 89 | 98 | |
| TSBH (Pass@1)Model=Llama3.2-3B2026.04 | 98 | 90 | |
| TSBH (Pass@5)Model=Llama3.2-3B2026.04 | 98 | 96 | |
| TSBH (Pass@1)Model=DeepSeek-7B2026.04 | 99 | 97 | |
| TSBH (Pass@5)Model=DeepSeek-7B2026.04 | 99 | 99 | |
| TSBH (Pass@1)Model=Qwen2.5-7B2026.04 | 99 | 100 | |
| TSBH (Pass@5)Model=Qwen2.5-7B2026.04 | 100 | 100 |