Prompt Injection Attack Evaluation on HarmBench
99.2ASPNeural-chat
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Neural-chatOpen-source=true2025.05 | 99.2 | 99.2 | |
| Stablelm2Open-source=true2025.05 | 98.9 | 97.8 | |
| MistralOpen-source=true2025.05 | 93.6 | 86.8 | |
| Starling-lmOpen-source=true2025.05 | 89.8 | 79.4 | |
| OpenchatOpen-source=true2025.05 | 86.1 | 85.9 | |
| VicunaOpen-source=true2025.05 | 81.9 | 81.8 | |
| Deepseek-r1Open-source=true2025.05 | 78.3 | 78.3 | |
| PhiOpen-source=true2025.05 | 72.7 | 72.7 | |
| Gemma2Open-source=true2025.05 | 48.8 | 48.8 | |
| Phi3Open-source=true2025.05 | 37 | 31.7 | |
| GemmaOpen-source=true2025.05 | 28 | 27.8 | |
| Gemma-2bOpen-source=true2025.05 | 21.3 | 21.2 | |
| Llama3Open-source=true2025.05 | 20.1 | 19.1 | |
| Llama2Open-source=true2025.05 | 17.8 | 17.8 |