Prompt Injection Detection on Hard-negative injection OOD
80.82Macro F1 ScoreTF–IDF only
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| TF–IDF onlyfeatures=TF-IDF2026.05 | 80.82 | 81.12 | 90.79 | 92.32 | 23.2 | 54.47 | 75.22 | |
| DeBERTa-basebackbone=DeBERTa, variant=base2026.05 | 73.13 | 74.73 | 86.41 | 87.84 | 35.2 | 59.98 | 64.31 |