Defense against Prompt Injection on DeepReview-13k (test)
0FPRPromptGuard
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PromptGuardAttack Type=GRPO-optimized2026.04 | 0 | — | 100 | |
| SafeReviewAttack Type=GRPO-optimized, Backbone=Qwen2026.04 | 38.2 | 0.409 | 33.7 | |
| SecAlignAttack Type=GRPO-optimized2026.04 | 42.2 | 0.343 | 35.1 | |
| System DefenseAttack Type=GRPO-optimized2026.04 | 43.6 | 0.365 | 29.2 |