About the possibility of using generative neural networks for malicious «high-quality»-purposes