Prompt Injection Attack on Direct Scenario
2.88ASRStruQ
Evaluation Results
| Method | Links | |
|---|---|---|
| StruQBackbone=Llama3-8B-Instruct, Attack Method=GCG2025.04 | 2.88 | |
| Robustness via ReferencingBackbone=Llama3-8B-Instruct, Attack Method=GCG2025.04 | 3.85 | |
| StruQBackbone=Llama3-8B-Instruct, Attack Method=AutoDAN2025.04 | 5.77 | |
| Robustness via ReferencingBackbone=Llama3-8B-Instruct, Attack Method=AutoDAN2025.04 | 6.25 | |
| Robustness via ReferencingBackbone=Qwen2-7B-Instruct, Attack Method=GCG2025.04 | 6.25 | |
| Robustness via ReferencingBackbone=Qwen2-7B-Instruct, Attack Method=AutoDAN2025.04 | 8.65 | |
| StruQBackbone=Qwen2-7B-Instruct, Attack Method=GCG2025.04 | 10.1 | |
| StruQBackbone=Qwen2-7B-Instruct, Attack Method=AutoDAN2025.04 | 14.42 | |
| SpotlightBackbone=Llama3-8B-Instruct, Attack Method=GCG2025.04 | 17.79 | |
| SandwichBackbone=Llama3-8B-Instruct, Attack Method=GCG2025.04 | 24.04 | |
| InstructionalBackbone=Llama3-8B-Instruct, Attack Method=GCG2025.04 | 25.48 | |
| SpotlightBackbone=Llama3-8B-Instruct, Attack Method=AutoDAN2025.04 | 25.48 | |
| ReminderBackbone=Llama3-8B-Instruct, Attack Method=GCG2025.04 | 27.88 | |
| SandwichBackbone=Llama3-8B-Instruct, Attack Method=AutoDAN2025.04 | 31.25 | |
| SandwichBackbone=Qwen2-7B-Instruct, Attack Method=GCG2025.04 | 32.21 | |
| InstructionalBackbone=Llama3-8B-Instruct, Attack Method=AutoDAN2025.04 | 39.9 | |
| SandwichBackbone=Qwen2-7B-Instruct, Attack Method=AutoDAN2025.04 | 40.38 | |
| SpotlightBackbone=Qwen2-7B-Instruct, Attack Method=GCG2025.04 | 41.83 | |
| ReminderBackbone=Llama3-8B-Instruct, Attack Method=AutoDAN2025.04 | 43.75 | |
| SpotlightBackbone=Qwen2-7B-Instruct, Attack Method=AutoDAN2025.04 | 47.12 | |
| InstructionalBackbone=Qwen2-7B-Instruct, Attack Method=GCG2025.04 | 57.69 | |
| NoneBackbone=Llama3-8B-Instruct, Attack Method=AutoDAN2025.04 | 60.58 | |
| ReminderBackbone=Qwen2-7B-Instruct, Attack Method=GCG2025.04 | 72.6 | |
| InstructionalBackbone=Qwen2-7B-Instruct, Attack Method=AutoDAN2025.04 | 72.6 | |
| ReminderBackbone=Qwen2-7B-Instruct, Attack Method=AutoDAN2025.04 | 85.58 | |
| NoneBackbone=Qwen2-7B-Instruct, Attack Method=GCG2025.04 | 88.94 | |
| NoneBackbone=Qwen2-7B-Instruct, Attack Method=AutoDAN2025.04 | 89.42 | |
| NoneBackbone=Llama3-8B-Instruct, Attack Method=GCG2025.04 | 92.27 |