Which you can use to train a better model to detect AI content.

I wonder whether some models fit this - well, you could certainly call it adversarial, and obviously generative - framework?