Jailbreak Defense on Safety Guardrail Evaluation Set
38Char Noise RobustnessQwen3Guard-Gen-8B
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Qwen3Guard-Gen-8B2026.03 | 38 | 24 | 39 | 31 | 100 | 26 | 83 | 53 | 44 | 2 | 100 | |
| ShieldGemma2-4B2026.03 | 24 | 9 | 9 | 12 | 56 | 8 | 39 | 13 | 11 | 3 | 21 | |
| LionGuard22026.03 | 16 | 10 | 21 | 14 | 52 | 53 | 45 | 44 | 52 | 13 | 28 | |
| LlamaGuard-3-8B2026.03 | 9 | 11 | 14 | 12 | 97 | 50 | 69 | 18 | 5 | 4 | 80 | |
| OpenAI Moderator2026.03 | 2 | 1 | 1 | 1 | 94 | 14 | 44 | 37 | 29 | 0 | 0 | |
| NemoGuard-8B2026.03 | 0 | 12 | 27 | 7 | 98 | 30 | 73 | 56 | 48 | 21 | 56 |