The neural network has learned to create compromising information on users of social networks.

Arstechnica researchers conducted an experiment. They created a fictional user named John, an elementary school teacher. And on the basis of his usual photo content, a real compromising evidence was generated. Here John is sitting half-naked in the classroom, taking pictures with a naked porn actress, walking in a clown costume, participating in an extremist group and serving time in prison. Experts used the Stable Diffusion neural network and the Dreambooth learning method, trained AI in just 1 hour and absolutely free.