Rerouting Detection on GSM8k (test)
1AccuracyRerouteGuard
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RerouteGuardAttack Scenario=Gray-box2026.01 | 1 | 1 | |
| RerouteGuardAttack Scenario=Box-free2026.01 | 1 | 1 | |
| RerouteGuardAttack Scenario=Average2026.01 | 1 | 1 | |
| RerouteGuardAttack Scenario=White-box2026.01 | 0.99 | 0.99 | |
| LlamaGuardAttack Scenario=Gray-box2026.01 | 0.55 | 0.52 | |
| PPLAttack Scenario=Gray-box2026.01 | 0.53 | 0.1 | |
| PPLAttack Scenario=White-box2026.01 | 0.52 | 0.06 | |
| LlamaGuardAttack Scenario=Average2026.01 | 0.52 | 0.5 | |
| PPLAttack Scenario=Average2026.01 | 0.51 | 0.05 | |
| LlamaGuardAttack Scenario=White-box2026.01 | 0.51 | 0.51 | |
| PPLAttack Scenario=Box-free2026.01 | 0.5 | 0 | |
| PromptGuardAttack Scenario=Gray-box2026.01 | 0.5 | 0.67 | |
| PromptGuardAttack Scenario=Box-free2026.01 | 0.5 | 0.67 | |
| PromptGuardAttack Scenario=White-box2026.01 | 0.5 | 0.67 | |
| PromptGuardAttack Scenario=Average2026.01 | 0.5 | 0.67 | |
| LlamaGuardAttack Scenario=Box-free2026.01 | 0.49 | 0.46 |