Comment Re:Any way for both sides to lose? (Score 1) 101
The issue is that with the proper set of prompts, the entire copyrighted work can be coaxed from the AI, which is distribution. Even more problematic is that numerous prompts can evince portions of the copyrighted work, portions large enough to qualify as copyright infringement.
If the AI kept it to themselves, they could train on as much copyrighted work as they want... but then, they would not be commercially available.
You can train the AI that you use on anything you want, copyrighted or not. An AI that is to be used commercially can not be trained on copyrighted works because the AI regurgitates portions, or all, of that copyrighted material.