OpenAI has published excerpts of amicus briefs filed in its lawsuit involving The New York Times, presenting a coordinated argument that fair use enables the training of large language models and other AI systems without undermining copyright protections. The materials include submissions from academics, industry associations, advocacy organizations, independent creators and the U.S. government, each framing model training as a transformative activity that serves broad public interests.
The briefs released by OpenAI converge on a common legal theory: using copyrighted works to train AI is a different, transformative purpose that does not function as a market substitute for the original works and therefore should qualify as fair use. One amici brief summarized that position by stressing that training relies on copyrighted material for analysis and learning rather than reproducing or replacing the works themselves, and warned against any judicial approach that would expand copyright beyond its established boundaries.
A coalition of library, accessibility and industry organizations emphasized the real-world consequences of narrowing fair use for AI training. Groups including the American Council of the Blind, Association of Research Libraries, American Library Association, Consumer Technology Association and Business Software Alliance argued that restrictions would have ripple effects well beyond AI companies. Their brief cautioned that limitations could hamper medical researchers, educators, people with disabilities, farmers, independent creators and many organizations that depend on third-party or open-source AI systems—stressing that fair use supports the ability to study diverse sources and to access tools that operationalize that learning.
Competition and market effects were another focal point. Authors Alliance contested theories that treat ordinary competitive success as an actionable harm under copyright law. Their brief warned that a concept of “market dilution” aimed at competitive pressure from noninfringing outputs would broaden copyright protection into domains it has not historically covered, eroding the principle that copyright prevents unauthorized copying of protected expression, not the commercial success of new works that compete for audiences.
Industry voices representing startups and developers framed model training as fundamentally different from the underlying training works. Organizations such as Engine and the Developers Alliance argued that startups build general-purpose tools using foundation models, not replicas of training texts. They described training as a pathway to adaptable technologies with purposes and outputs distinct from the source material used for learning.
Several submissions urged courts to weigh national and research benefits when assessing market effects under the fourth fair-use factor. The Foundation for American Innovation highlighted the potential gains in analysis, software and capabilities that flow from model training, and recommended considering national-security and productivity values as part of the public-benefit assessment.
Academic perspectives in the briefs argued for maintaining incentives for creating tools that learn from content. Economist Joshua S. Gans cautioned that treating every displaced sale as an actionable copyright violation could weaken incentives to develop technologies that depend on learning from existing works. Professor Matthew Sag characterized the defendants’ use of plaintiffs’ works for training as highly transformative, noting that, in realistic conditions, model outputs are not equivalent to the intermediate copies made during the training process.
Independent creators also filed in support of preserving the scope of fair use for AI training. Michael Wall and Sarah Rose Siskind described how they rely on copyright protections while using AI for business tasks, ideation and expanding creative output. They said AI tools help them work more efficiently and connect with new clients, and have become integral to their practices.
Public interest advocates pointed to historical precedents in which fair use protected tools that navigate large information collections—such as image and full-text search—and the production of accessible formats for readers with print disabilities. A separate filing by Ron Suskind and Professor Brian L. Frye argued that OpenAI’s use of newspaper articles for training did not prevent news organizations from selling access to their articles.
Notably, the United States government submitted a brief urging caution against narrowing fair-use doctrine to exclude large language model training. The government warned such a contraction would conflict with foundational copyright principles and could impede scientific and creative progress, noting that researchers and publications already use LLMs for journalistic and research activities.
OpenAI’s collection of excerpts points readers to the public court docket for full filings and includes background material on the case. By making these amici arguments available, the company has foregrounded the legal and policy stakes of how courts interpret fair use in the context of rapidly evolving AI technologies.
As the litigation proceeds, the variety of amici briefs underscores that the question of fair use and AI training touches many sectors—research, education, accessibility, innovation and creative practice—each warning that an overly restrictive reading could produce broad and unintended consequences. The materials posted by OpenAI leave little doubt that multiple stakeholders view the preservation of established fair-use principles as critical to ongoing technological and societal progress.
Source: Read the original source
Leave a Reply