Prompt Injection Attack Evaluation on SAP10
99.4Attack Success Percentage (ASP)Neural-chat
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Neural-chatOpen-source=true2025.05 | 99.4 | 99.2 | |
| Stablelm2Open-source=true2025.05 | 98.5 | 98.3 | |
| MistralOpen-source=true2025.05 | 96.5 | 93.3 | |
| OpenchatOpen-source=true2025.05 | 96.3 | 94.6 | |
| Starling-lmOpen-source=true2025.05 | 88.5 | 82.1 | |
| PhiOpen-source=true2025.05 | 82.1 | 80.8 | |
| VicunaOpen-source=true2025.05 | 72.9 | 72.1 | |
| Deepseek-r1Open-source=true2025.05 | 67.3 | 65.4 | |
| Gemma2Open-source=true2025.05 | 62.1 | 56.7 | |
| Phi3Open-source=true2025.05 | 31 | 29.6 | |
| Gemma-2bOpen-source=true2025.05 | 17.1 | 16.2 | |
| Llama2Open-source=true2025.05 | 11 | 4.2 | |
| GemmaOpen-source=true2025.05 | 2.7 | 2.5 | |
| Llama3Open-source=true2025.05 | 0.8 | 0 |