Overview
- The Department of Justice filed a statement of interest Tuesday arguing that ingesting copyrighted text to train large language models is “extraordinarily transformative” and generally qualifies as fair use.
- DOJ lawyers told the Manhattan federal court that limiting training would slow scientific progress, weaken U.S. firms against foreign rivals, and threaten national security by constraining tools used for analysis and intelligence work.
- The New York Times and other plaintiffs sharply criticized the filing, saying the government is siding with large AI companies and that creators should be paid for content used to build models.
- The brief separates the legality of the training process from harms caused by model outputs and urged the court not to treat output uses as dispositive of fair use for training.
- The filing is non-binding but could influence many pending suits because the NYT case is one of several that could set precedent after mixed rulings and settlements in prior AI copyright litigation.