I understand why they’re doing it, but I don’t like the asymmetry, especially as we (the reviewers) are then having to read AI-generated text, with associated waste of our time (which we are not compensated for).

I agree it's causing all sorts of problems but I can't see that saying it's not allowed would make any difference - impossible to police, and only the weak uses of it are obvious - my instinct is that most use by academics would be undectable. Not sure what a good policy would/could look like.