System Prompt Extraction on System Prompt Extraction Set In-merge
75Attack Success RateRogueMerge
Evaluation Results
| Method | Links | |
|---|---|---|
| RogueMergeMerging Algorithm=Model Breadcrumbs (MB)2026.06 | 75 | |
| RogueMergeMerging Algorithm=DARE2026.06 | 64 | |
| CleanMerging Algorithm=DARE2026.06 | 11 | |
| CleanMerging Algorithm=Model Breadcrumbs (MB)2026.06 | 7 |