System Prompt Extraction on System Prompt Extraction Set Out-of-merge
77.5Attack Success RateRogueMerge
Evaluation Results
| Method | Links | |
|---|---|---|
| RogueMergeMerging Algorithm=Model Breadcrumbs (MB)2026.06 | 77.5 | |
| RogueMergeMerging Algorithm=DARE2026.06 | 74 | |
| CleanMerging Algorithm=Model Breadcrumbs (MB)2026.06 | 9.5 | |
| CleanMerging Algorithm=DARE2026.06 | 7.5 |