Deception Detection on 7 non-control benchmark families Mean across targets
91.6AUROCSTATEWITNESS
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| STATEWITNESS2026.06 | 91.6 | 32.6 | 62.8 | |
| Best black-boxSelection=Best configuration2026.06 | 82.1 | 50.2 | 62.2 | |
| Best linear probeSelection=Best configuration2026.06 | 73.3 | 37.9 | 47.4 |