In dialects of Levantine Arabic, "human being" is banīᐣādam بَني آدَم, lit. "sons of Adam," but it's been reanalyzed as a single word: the feminine is banīᐣādmeh بَني آدمة and the plural banīᐣādmīn بَني آدمين, with the feminine/plural markers on "Adam" instead of "son."

One possible reason that it was reanalyzed is that banī is a plural genitive construct, which isn't productive in Levantine, and because banīn بَنين is no longer the unmarked plural of ᐣibn إبن "son," having been replaced by a suppletive form wlād ولاد.

At the same time, increased knowledge of Classical/Standard Arabic makes it clear to all speakers that this "word" is actually a compound (which is why I elected to write it as such in Arabic). Still, this knowledge hasn't affected its morphological behavior.

This is quite different from almost every other rebracketing example that I'm aware of, where the original word boundary is no longer apparent. I'm curious if there are any other cases like this where the rebracketing process has been partially or fully reversed.