Jailbreak Attack on LLM-LAT Out-of-merge
72.3Attack Success RateRogueMerge
Evaluation Results
| Method | Links | |
|---|---|---|
| RogueMergeMerging Algorithm=Model Breadcrumbs (MB)2026.06 | 72.3 | |
| RogueMergeMerging Algorithm=DARE2026.06 | 66.3 | |
| CleanMerging Algorithm=Model Breadcrumbs (MB)2026.06 | 32 | |
| CleanMerging Algorithm=DARE2026.06 | 26.7 |