Fair use and training data

DOJ asks the court to judge LLM training separately from model outputs

The 20-page filing concedes an output can infringe on its own while asking that training be judged apart from it. If the judge agrees, the risk builders carry moves into the serving path, where it can be tested.