Prompt Injection Detection on Standard Injection Qualifire (OOD)
72.96Macro F1Semantic+IBVS Boost
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Semantic+IBVS Boostvariant=Boost2026.05 | 72.96 | 73.1 | 76.79 | 80.27 | 6.25 | 23.56 | 40.98 | |
| Semantic+IBVS BoostRegime=Std-Inj. OOD2026.05 | 72.96 | 73.1 | 76.79 | 80.27 | 6.25 | 23.56 | 40.98 | |
| LLM Guard PromptInjectionRegime=Std-Inj. OOD2026.05 | 72.35 | 72.53 | 81.31 | 82.58 | 19.02 | 33.32 | 44.56 | |
| RoBERTa+IBVS Structuredvariant=Structured2026.05 | 63.66 | 66.76 | 79.56 | 81.17 | 11.79 | 30.54 | 47.57 | |
| RoBERTa+IBVS StructuredRegime=Std-Inj. OOD2026.05 | 63.66 | 66.76 | 79.56 | 81.17 | 11.79 | 30.54 | 47.57 |