Comment by staticassertion

The attack surface is the parser. The ability to access it is arbitrary. I can't build a threat model beyond that for any specific case, but in the case of a text messaging app I absolutely expect "attacker can text you" to be in your threat model.