Adversarial Attack on AdvBench Hijacked (test)
0CHRFPFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FPFTModel=DeepSeek-7B2026.04 | 0 | 20 | |
| OriginalModel=Llama3.2-3B2026.04 | 1 | 1 | |
| AutoRANModel=Qwen2.5-7B2026.04 | 8 | 19 | |
| AutoRANModel=DeepSeek-7B2026.04 | 10 | 14 | |
| AutoRANModel=Llama3.2-3B2026.04 | 20 | 22 | |
| CoT HijackingModel=Llama3.2-3B2026.04 | 26 | 11 | |
| OriginalModel=Qwen2.5-7B2026.04 | 30 | 6 | |
| OriginalModel=DeepSeek-7B2026.04 | 32 | 19 | |
| CoT HijackingModel=DeepSeek-7B2026.04 | 42 | 33 | |
| Chain-of-LureModel=Llama3.2-3B2026.04 | 46 | 36 | |
| CoT HijackingModel=Qwen2.5-7B2026.04 | 47 | 32 | |
| Chain-of-LureModel=DeepSeek-7B2026.04 | 52 | 47 | |
| Chain-of-LureModel=Qwen2.5-7B2026.04 | 53 | 45 | |
| H-CoTModel=Qwen2.5-7B2026.04 | 57 | 15 | |
| H-CoTModel=Llama3.2-3B2026.04 | 57 | 5 | |
| H-CoTModel=DeepSeek-7B2026.04 | 59 | 33 | |
| FPFTModel=Llama3.2-3B2026.04 | 90 | 82 | |
| LRFTModel=Llama3.2-3B2026.04 | 93 | 99 | |
| LRFTModel=DeepSeek-7B2026.04 | 94 | 99 | |
| LRFTModel=Qwen2.5-7B2026.04 | 94 | 99 | |
| TSBH (Pass@5)Model=DeepSeek-7B2026.04 | 98 | 99 | |
| FPFTModel=Qwen2.5-7B2026.04 | 99 | 98 | |
| TSBH (Pass@1)Model=Qwen2.5-7B2026.04 | 99 | 99 | |
| TSBH (Pass@5)Model=Qwen2.5-7B2026.04 | 99 | 99 | |
| TSBH (Pass@1)Model=DeepSeek-7B2026.04 | 100 | 99 | |
| TSBH (Pass@1)Model=Llama3.2-3B2026.04 | 100 | 99 | |
| TSBH (Pass@5)Model=Llama3.2-3B2026.04 | 100 | 100 |