This page collects excerpts from amicus briefs filed in the copyright lawsuit brought by The New York Times over OpenAI’s use of its articles to train generative language models. The introduction argues that fair use has long supported innovation, from medical and scientific breakthroughs to new technologies that benefit authors and journalists, and that the Times’ suit asks the Court to ignore a half century of fair-use history for an “undeserved payday.” Every excerpt below supports the position that AI training qualifies as fair use.

Several briefs make the core legal argument that training is transformative and does not substitute for the originals. Matthew Sag writes that, on the record, the defendants’ use is highly transformative because under realistic conditions model outputs are in no relevant way similar to the invisible intermediate copies made during training. The Engine and Developers Alliance argues that training a foundation model is not intended to reproduce the works it learns; it is building a general-purpose tool for users to adapt, so the purpose bears no resemblance to the purposes of the original works. The American Council of the Blind, Association of Research Libraries, American Library Association, Consumer Technology Association & Business Software Alliance add that the case affects medical researchers, patients, educators, people with disabilities, farmers, independent creators, and open source AI users, not just LLM developers.

The briefs also push back on the idea that AI outputs cause legally cognizable market harm. Authors Alliance asks the Court to reject “market dilution”—aggregate competitive pressure from noninfringing outputs that occupy the same subject matter or audience—as cognizable harm under 17 U.S.C. § 107(4). New works can reduce the sales or visibility of older works without copying protected expression, and copyright law deliberately leaves that competition outside the owner’s control. Joshua S. Gans makes a related point: copyright supports both the incentive to create content and the incentive to build tools that learn from content, and treating every displaced sale as actionable would weaken the second without reliably protecting the first. Ron Suskind and Brian L. Frye argue that OpenAI’s use of newspaper articles did not harm the plaintiffs’ reasonable commercial expectations because the Times retains and continues to exercise the exclusive right to sell readers access to its articles.

Several submissions focus on creators and public access. Comedian/writer Sarah Rose Siskind and musician Michael Wall describe using AI to handle administrative and operational work, reduce costs, and develop ideas, while maintaining that AI does not displace their creativity; Siskind says she could not run her business without AI at this point. Kevin Kelly describes AI as a transformative medium that lets authors create new texts and reason, and warns that a ruling for the plaintiffs would hinder new modes of thought and the creation of original works. Public Knowledge compares AI training to prior fair use protections for image search, Google Books, HathiTrust’s accessible digital books for readers with print disabilities, and free online publication of technical standards incorporated into law, arguing that fair use will ensure AI’s benefits accrue to the public rather than a privileged few.

The United States Government brief warns that the Times’ request to narrow fair use to exclude LLM training would be inconsistent with basic copyright law principles and would severely hamper “the Progress of Science and useful Arts.” It notes that publications, including independent and startup media, are already using LLMs for journalistic work, and that LLMs are helping researchers achieve major breakthroughs. The Foundation for American Innovation adds that AI-enabled capabilities may carry national-security value, which should factor into the market-effects assessment. Taken together, the excerpts ask the Court to treat generative AI training as fair use and to reject the Times’ attempt to expand copyright liability beyond settled limits.