> There used to be the problem with authors adding new co-authors after acceptance, aka "selling seats".

Top AI conferences allow that? That's insane.