For policy setting that's the wrong question. The real ones are: how likely is it to happen, how much would training/comms change that, and how acceptable is that risk. Their answers seem to have been: very, not enough, it's not.

Where the problem is and how to change it is an interesting topic, but likely irrelevant for the decision.

For Apple it's a risk management problem.

Hence why they are banning it.

No, not, if you can't change the users.

If apple hosts it's one chatgpt internally it is neither.