It would make infinitely more sense to get around it contractually. To use the service, the prompter agrees to take liability for the output. More like upscal

It would make infinitely more sense to get around it contractually. To use the service, the prompter agrees to take liability for the output. More like upscaling. Should a service be liable for upscaling an illegal picture, that they "produced?" If LLM output is reverse compression, it is upscaling of the original prompt. (That said, I dont think your example of a model compressing the internet really invokes 230, because the model creator is the one grabbing the content, not running a service that other people post content to.)