Harmful Request Defense on authority probes
75Block RateCognitive Firewall
Evaluation Results
| Method | Links | |
|---|---|---|
| Cognitive FirewallDefense Type=Trajectory-aware firewalls, Target Model=mistral, Judge Model=qwen3.5:122b2026.07 | 75 | |
| Granite Guardian 8BDefense Type=Per-message guard models, Target Model=mistral, Judge Model=qwen3.5:122b2026.07 | 56 | |
| THRD-styleDefense Type=Trajectory-aware firewalls, Target Model=mistral, Judge Model=qwen3.5:122b2026.07 | 31 | |
| TCA-styleDefense Type=Trajectory-aware firewalls, Target Model=mistral, Judge Model=qwen3.5:122b2026.07 | 31 | |
| ShieldGemma 9BDefense Type=Per-message guard models, Target Model=mistral, Judge Model=qwen3.5:122b2026.07 | 12 | |
| Llama Guard 3 8BDefense Type=Per-message guard models, Target Model=mistral, Judge Model=qwen3.5:122b2026.07 | 6 | |
| none (undefended)Defense Type=undefended, Target Model=mistral, Judge Model=qwen3.5:122b2026.07 | 0 |