The caster of magic fang typically can pick only a natural weapon that the subject already possesses to be affected by the spell…
What usually dictates whether a creature is or isn't a valid subject for a spell is a spell's Target entry (see Magic on Aiming a Spell on Target or Targets for details). The spell magic fang has the entry Target: Living creature touched, but, upon casting the spell and subsequently touching the target, the caster picks "one natural weapon or unarmed strike of the subject [to receive] a +1 enhancement bonus on attack and damage rolls" (emphasis mine). Picking something that a creature doesn't have as the subject of an effect usually causes the effect to fail.
Thus I suspect that most GMs will rule that, while the spell's Target entry does mandate only that the subject be a living creature the caster touches, the spell's description increases the specificity of that Target entry to include "one natural weapon or unarmed strike of the subject," and if the subject does not possesses the natural weapon when the spell is cast, the spell fails.
…But this GM would likely let it slide anyway
However, the magic fang spell's Target entry is sufficient for this GM to rule that the spell doesn't outright fail if the caster picks a natural weapon the subject doesn't have to gain the spell's effect. Instead, this GM would rule that until the creature gains the picked natural weapon the subject's magic fang spell effect is present but, essentially, dormant, its duration counting down normally. This GM's ruling is—perhaps excessively—permissive, though. That is, if this GM can find any reading of a spell that allows a spell to work instead of to fail outright, this GM usually gravitates toward the reading that allows the spell to work. But were a GM to rule that spell magic fang functions in the traditional fashion—capable of affecting only an existing natural weapon—this player wouldn't flip the table and leave the campaign .
