AI Detector Evasion on MAGE (evaluation set)
91.3ASR (τ=0.5)BERTScore Linear
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| BERTScore Linearlambda=0.12026.05 | 91.3 | 0.883 | 0.805 | — | 99.4 | 0.05 | |
| BERTScore 0Optimization Mode=Detector Evasion Only2026.05 | 89.5 | 0.903 | 0.798 | — | 99.5 | 0.028 | |
| BERTScore Linearlambda=0.32026.05 | 76.8 | 0.758 | 0.834 | — | 95.3 | 0.081 | |
| DEPO2026.05 | 72.5 | 0.716 | 0.857 | — | 95.7 | 0.14 | |
| BERTScore Linearlambda=0.52026.05 | 34.7 | 0.348 | 0.837 | — | 66.8 | 0.344 | |
| Adv. Paraphrase2026.05 | 9.4 | 0.073 | 0.826 | — | 48.9 | 0.453 | |
| AuthorMist2026.05 | 5.9 | 0.07 | 0.81 | — | 32.1 | 0.466 | |
| Direct Paraphrasing2026.05 | 3 | 0.04 | 0.854 | — | 27.6 | 0.476 | |
| StealthRL2026.05 | 2.1 | 0.045 | 0.613 | — | 78.9 | 0.459 | |
| BERTScore 1Optimization Mode=Semantic Only2026.05 | 0.8 | 0.01 | 0.993 | — | 8.1 | 0.494 | |
| Original2026.05 | 0.7 | — | — | — | 7.8 | 0.494 | |
| SilverSpeak2026.05 | 0.5 | 0.028 | 0.338 | — | 84.9 | 0.469 |