Indirect Prompt Injection Detection on Jfleg RTE
99.4AccuracyRENNERVATE
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| RENNERVATEModel=LLaMA32025.12 | 99.4 | 100 | 0.2 | |
| RENNERVATEModel=LLaMA32025.12 | 99.35 | 0.4 | 90 | |
| RENNERVATEModel=ChatGLM2025.12 | 98.55 | 2.7 | 20 | |
| RENNERVATEModel=ChatGLM2025.12 | 96.9 | 60 | 5.6 | |
| RENNERVATEModel=Dolly2025.12 | 96.65 | 3.3 | 340 | |
| RENNERVATEModel=Dolly2025.12 | 96.2 | 560 | 2 | |
| RENNERVATEModel=LLaMA22025.12 | 96 | 800 | 0 | |
| RENNERVATEModel=LLaMA22025.12 | 94.25 | 11.5 | 0 | |
| RENNERVATEModel=Falcon2025.12 | 94.2 | 2 | 960 | |
| RENNERVATEModel=Falcon2025.12 | 82.2 | 2,150 | 14.1 |