Prompt Injection Attack Evaluation on AdvBench
99.9ASPMistral
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MistralOpen-source=true2025.05 | 99.9 | 99.9 | |
| Neural-chatOpen-source=true2025.05 | 99.5 | 99.5 | |
| OpenchatOpen-source=true2025.05 | 94 | 92 | |
| Stablelm2Open-source=true2025.05 | 93.1 | 93.1 | |
| Starling-lmOpen-source=true2025.05 | 81.3 | 76.6 | |
| VicunaOpen-source=true2025.05 | 75 | 73.5 | |
| PhiOpen-source=true2025.05 | 68.6 | 57.2 | |
| Deepseek-r1Open-source=true2025.05 | 60.5 | 60.5 | |
| Gemma2Open-source=true2025.05 | 55.7 | 35.1 | |
| Phi3Open-source=true2025.05 | 32 | 22.9 | |
| Llama2Open-source=true2025.05 | 12.2 | 0.2 | |
| GemmaOpen-source=true2025.05 | 6.2 | 1.6 | |
| Llama3Open-source=true2025.05 | 1.4 | 0.4 | |
| Gemma-2bOpen-source=true2025.05 | 1.4 | 1.2 |