Rerouting Detection on MT-Bench (test)
100AccuracyRerouteGuard
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RerouteGuardAttack Scenario=Gray-box2026.01 | 100 | 100 | |
| RerouteGuardAttack Scenario=Box-free2026.01 | 100 | 100 | |
| RerouteGuardAttack Scenario=White-box2026.01 | 100 | 100 | |
| RerouteGuardAttack Scenario=Average2026.01 | 100 | 100 | |
| PPLAttack Scenario=White-box2026.01 | 72 | 62 | |
| PPLAttack Scenario=Gray-box2026.01 | 68 | 53 | |
| PPLAttack Scenario=Average2026.01 | 63 | 38 | |
| LlamaGuardAttack Scenario=Box-free2026.01 | 58 | 57 | |
| LlamaGuardAttack Scenario=Average2026.01 | 51 | 50 | |
| PPLAttack Scenario=Box-free2026.01 | 50 | 0 | |
| PromptGuardAttack Scenario=Box-free2026.01 | 50 | 67 | |
| LlamaGuardAttack Scenario=Gray-box2026.01 | 48 | 45 | |
| PromptGuardAttack Scenario=Gray-box2026.01 | 47 | 64 | |
| PromptGuardAttack Scenario=Average2026.01 | 47 | 64 | |
| LlamaGuardAttack Scenario=White-box2026.01 | 47 | 47 | |
| PromptGuardAttack Scenario=White-box2026.01 | 44 | 62 |