I've been waiting for this. My friends often talk about how scary deep fakes are, but I'm not really too worried at this point. I've heard that though generated images hold up well to cursory human observation they break down under statistical analysis.
I think it's not a vicious circle going nowhere, but a vicious circle that leads to accelerating advances in AI. We will learn loads from both ends of this kind of effort and AI on both sides will be more efficacious because of it. We may even learn something insightful about perception, creativity, or computing more generally.