Misalignment Detection on Covert code sabotage Bloom behaviors (in-distribution)
0.959AUROC18-indicator ensemble
Evaluation Results
| Method | Links | |
|---|---|---|
| 18-indicator ensembleLayer=272026.06 | 0.959 | |
| 18-indicator ensembleLayer=292026.06 | 0.956 | |
| RepE baselineLayer=272026.06 | 0.943 | |
| RepE baselineLayer=292026.06 | 0.904 |