Indirect Prompt Injection Defense on BIPIA Average EN/BN
70.3BUQwen3-8B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-8BDefense=No Defense (ND)2026.05 | 70.3 | 15 | 96.8 | |
| Qwen2.5-7B-InstructDefense=MIPIAD (D)2026.05 | 68.4 | 35 | 92.4 | |
| Qwen2.5-7B-InstructDefense=No Defense (ND)2026.05 | 68.2 | 31 | 92 | |
| Qwen3-8BDefense=MIPIAD (D)2026.05 | 67.6 | 26 | 96.4 | |
| Qwen3.5-9BDefense=No Defense (ND)2026.05 | 66.4 | 19 | 91.2 | |
| Qwen3.5-9BDefense=MIPIAD (D)2026.05 | 64.9 | 2 | 78 | |
| Gemma-3-12B-ITDefense=MIPIAD (D)2026.05 | 57.1 | 23 | 85.5 | |
| TigerLLM-9B-ITDefense=MIPIAD (D)2026.05 | 57.1 | 25 | 88.1 | |
| Macro avg.Defense=No Defense (ND)2026.05 | 53.5 | 16 | 90 | |
| Macro avg.Defense=MIPIAD (D)2026.05 | 53.4 | 19 | 89.4 | |
| TigerLLM-9B-ITDefense=No Defense (ND)2026.05 | 53.3 | 10 | 86.7 | |
| Gemma-3-12B-ITDefense=No Defense (ND)2026.05 | 52.5 | 11 | 86.3 | |
| BanglaLLaMA-3-8BDefense=No Defense (ND)2026.05 | 52 | 19 | 84.9 | |
| BanglaLLaMA-3-8BDefense=MIPIAD (D)2026.05 | 43.3 | 15 | 90.6 | |
| BLOOMZ-7B1-MTDefense=MIPIAD (D)2026.05 | 15.3 | 10 | 94.7 | |
| BLOOMZ-7B1-MTDefense=No Defense (ND)2026.05 | 11.7 | 6 | 92.1 |