Model Stealing Attacks on Mistral
0.985BERT ScoreSAFESEAL
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SAFESEALSetting=Watermark2026.05 | 0.985 | 97.4 | — | — | |
| LWSetting=Watermark2026.05 | 0.982 | — | 28 | 2.3 | |
| TWSetting=Watermark2026.05 | 0.961 | 95.6 | — | — | |
| SAFESEALSetting=Adv.(20k)2026.05 | 0.88 | 61.9 | — | — | |
| TWSetting=Adv.(20k)2026.05 | 0.879 | 1 | — | — | |
| SAFESEALSetting=Adv.(10k)2026.05 | 0.878 | 50 | — | — | |
| LWSetting=Adv.(20k)2026.05 | 0.877 | — | 11 | 1 | |
| TWSetting=Adv.(10k)2026.05 | 0.876 | 0.5 | — | — | |
| LWSetting=Adv.(10k)2026.05 | 0.866 | — | 1 | 1 |