The signing of the update prevents it.
I mean, technically there is an attack vector there, but it's oddly specific, and since the attacker needs physical access anyways, you've got other problems (i.e. someone got into your home).
Of course, having phones be updated before users use them is also super important. So maybe what’s going on here is something that only works when a phone hasn’t been associated to a user yet.
It seems to me the threat model for consumers is the same either way. The box itself doesn't offer any meaningful tamper-evidence, so a sophisticated attacker would just re-shrinkwrap everything if they couldn't update them this way.