Benign Prompt Refusal on XS-Test and OR-Bench-Hard
9.61FPRLlama-Guard-3
Evaluation Results
| Method | Links | |
|---|---|---|
| Llama-Guard-3Backbone=Llama-3-8B-Instruct2026.06 | 9.61 | |
| NeuroArmor-ReAnchorBackbone=Llama-3-8B-Instruct2026.06 | 22.05 | |
| VanillaBackbone=Llama-3-8B-Instruct2026.06 | 30.26 | |
| Runtime Intervention BaselineBackbone=Llama-3-8B-Instruct2026.06 | 33.4 | |
| Direct Safety SteeringBackbone=Llama-3-8B-Instruct2026.06 | 33.54 |