List your company

Jailbreak Attack on LLM-LAT Out-of-merge

72.3Attack Success RateRogueMerge
24.87637.18849.561.812Jun 2, 2026

Evaluation results

MethodLinks
RogueMergeMerging Algorithm=Model Breadcrumbs (MB)2026.0672.3
RogueMergeMerging Algorithm=DARE2026.0666.3
CleanMerging Algorithm=Model Breadcrumbs (MB)2026.0632
CleanMerging Algorithm=DARE2026.0626.7