I think the main thing that throws it into question is when you get the software that sends the data to the service and the service from the same people (in this case apple). You're already trusting them with your data, and a fancy HE scheme doesn't change that. They can update their software and start sending everything in plain text and you wouldn't even realise they'd done it.

FHE is plausibly most useful when you trust the source of the client code but want to use the compute resource of an organisation you don't want to have to trust.