Attack Detection on OverThink N = 100
100TPR (OverThink N=100)Token Budget
Evaluation results
| Method | Links | |
|---|---|---|
| Token BudgetBackbone Model=DS-R1-Qwen-7B2026.06 | 100 | |
| Compression-Ratio ProxyBackbone Model=DS-R1-Qwen-7B2026.06 | 100 | |
| Output-Length AnomalyBackbone Model=DS-R1-Qwen-7B2026.06 | 100 | |
| QDMBackbone Model=DS-R1-Qwen-7B2026.06 | 100 | |
| RECURGUARDBackbone Model=DS-R1-Qwen-7B2026.06 | 99 | |
| Answer-Absence ProxyBackbone Model=DS-R1-Qwen-7B2026.06 | 98 |