> I'm not sure current genAI models do these sorts of things.

You might be right here. Two points though - first, we don't know if current AI is actually incapable of something in particular; we didn't find this, didn't prove it. Second, we might have a different AI approach, which would actually be capable of these things you mention. To me, it's way too early to dismiss AIs - at least in principle - regarding all of this.