Preprint
Aug 2026
A Single Suffix to Break Them All: Basin-Aware Jailbreaks for Merged Model Families
This work proposes Basin-Aware Jailbreak (BAJ), which formulates jailbreak generation as a min--max optimization over the merging space to produce transferable adversarial suffixes across merged model families.
Yu Zhe, Yiting Tan, Jun-Hao Wei et al.
· 0 citations