Access Control on 12 template–attack datasets (test)
100PrivacyHarness-MU
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Harness-MUModel=Deepseek, Access Type=Open-Weights Models2026.06 | 100 | 98.9 | 99.3 | |
| Muses-BenchModel=Qwen, Access Type=Open-Weights Models2026.06 | 100 | 71.5 | 75.3 | |
| Harness-MUModel=Qwen, Access Type=Open-Weights Models2026.06 | 100 | 99.2 | 99.3 | |
| Harness-MUModel=Gemini, Access Type=Proprietary Models2026.06 | 100 | 99.1 | 99.4 | |
| Harness-MUModel=GPT, Access Type=Proprietary Models2026.06 | 100 | 58.9 | 63.7 | |
| Muses-BenchModel=GPT, Access Type=Proprietary Models2026.06 | 99.5 | 25.7 | 28.4 | |
| Muses-BenchModel=Gemini, Access Type=Proprietary Models2026.06 | 94 | 69.7 | 67.8 | |
| Muses-BenchModel=Deepseek, Access Type=Open-Weights Models2026.06 | 92.6 | 59.7 | 56.6 |