Harmful Request Defense on jbb
0Attack Success Rate (ASR)Granite Guardian 8B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Granite Guardian 8BDefense Type=Per-message guard models, Target Model=mistral, Judge Model=qwen3.5:122b2026.07 | 0 | 100 | |
| THRD-styleDefense Type=Trajectory-aware firewalls, Target Model=mistral, Judge Model=qwen3.5:122b2026.07 | 0 | 99 | |
| TCA-styleDefense Type=Trajectory-aware firewalls, Target Model=mistral, Judge Model=qwen3.5:122b2026.07 | 0 | 98 | |
| Llama Guard 3 8BDefense Type=Per-message guard models, Target Model=mistral, Judge Model=qwen3.5:122b2026.07 | 1 | 96 | |
| Cognitive FirewallDefense Type=Trajectory-aware firewalls, Target Model=mistral, Judge Model=qwen3.5:122b2026.07 | 1 | 92 | |
| ShieldGemma 9BDefense Type=Per-message guard models, Target Model=mistral, Judge Model=qwen3.5:122b2026.07 | 26 | 55 | |
| none (undefended)Defense Type=undefended, Target Model=mistral, Judge Model=qwen3.5:122b2026.07 | 45 | 0 |