Prompt Injection Detection on PaperGuard (test)
95AccuracyChunk-based
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Chunk-basedMethod Category=Embedding Search2026.06 | 95 | 92.86 | 0 | 7.14 | |
| LLM-as-JudgeMethod Category=Proprietary models2026.06 | 66.7 | 100 | 100 | 0 | |
| Embedding ClassifierMethod Category=Trained Classifier2026.06 | 64.5 | 0 | 12 | 100 | |
| BERT ClassifierMethod Category=Trained Classifier2026.06 | 38.5 | 0 | 35 | 100 | |
| Moderation APIMethod Category=Proprietary models2026.06 | 33.3 | 0 | 0 | 100 |