Just a point of clarity, an AI model capable of generating csam doesn’t necessarily have to be trained on csam.
Why is that? The whole point of generative AI is that it can combine concepts.
You train it on the concept of a chair using only red chairs. You train it on the color red, and the color blue. With this info and some repetition, you can have it output a blue chair.
The same applies to any other concepts. Larger, smaller, older, younger. Man, boy, woman, girl, clothed, nude, etc. You can train them each individually, gradually, and generate things that then combine these concepts.
Obviously this is harder than just using training data of what you want. It’s slower, it takes more effort, and results are inconsistent, but they are results. And then, you curate the most viable of the images created this way to train a new and refined model.
