It will get orders of magnitude worse once all this can be fully automated by AI (including a completely realistic video or voice call by the grandson where he talks about killing the pregnant woman).
One needs to provide clear instructions for a single good scam when explicitly asked to do so, the other would need to provide good defenses to any number of scams and be more or less self-activated.
There's also the matter of risks entailed by having large swathes of the population increasingly under the direct supervision of artificial minds (controlled by who?).
I suspect something like this will happen though, and that one of the key uses of artificial intelligence will be as both an offensive and a defensive weapon for deceit and manipulation. Indeed the offensive part would seem to be a natural continuation of the enormous resources poured into ml-driven engagement maximization and the quest for sufficiently politically correct AIs. As an example ChatGTP is evidently explicitly designed to lie about some things, including its own capabilities. I'm not sure what, if anything, currently exists on the defensive front.
Yeah for voice calls nothing yet though the likes of Google/Apple could probably do something. Against your point that bad AI is easier, you have that Google/Apple have more resources than the scammers.