EFF files amicus brief warning fair use curbs would harm public

EFF filed an amicus (friend-of-the-court) brief on Oct. 30, 2026, in re OpenAI contending that curbing fair use in generative AI training would damage the public interest.

The case combines 12 lawsuits against OpenAI, including The New York Times’s 2023 suit against OpenAI and Microsoft. Judge Sidney H. Stein has not ruled on whether training copyrighted content for large language models is fair use or on whether to accept the brief. The brief explains how the court’s fair use ruling could affect people worldwide, according to EFF. Publishers and creators are suing large AI companies over training generative AI models on copyrighted works.

EFF warns that if plaintiffs prevail, courts would have to redraw copyright law in ways that harm everyone except powerful media and tech companies. Plaintiffs seek to limit fair use rights for everyone, thwarting copyright’s basic purpose, according to EFF. Plaintiffs argue even transformative AI training is unfair because models might produce new works that compete with their own, and that scraping or downloading data without authorization was independently infringing. EFF says this would prevent transformative uses that help everyone.

Researchers, journalists, and watchdog groups gather information using approaches like those of AI companies, but they do so for the public good and frequently to hold Big Tech accountable. Journalists and nonprofit researchers have relied on crawling or scraping to scrutinize tech companies for potentially anticompetitive behavior and to research misinformation and hateful content on X. The data collection methods of these researchers closely match those of AI giants, but the researchers lack the companies’ financial resources, EFF says.

Obligating users to show lawful acquisition of copyrighted material would endanger routine internet activity, EFF argues. This rule would create legal uncertainty over billions of everyday copies made when people open e-mails, click links, visit websites, share photos, stream music, or watch videos. These routine activities all depend on intermediate copying, the brief states. Fair use allows internet use without forcing people to trace the source of every image, video, or article they click and confirm they have cleared rights, EFF says.

Mandated permission for each work would benefit the largest companies over small rivals and slow academic research, EFF says. Professor Lawrence Lessig recently warned that if every training use requires a license, only a few tech giants could afford the costs. “If all training requires a license, then none but a handful of tech giants will be able to afford the costs,” Lessig said. Lessig said this is not because authors are greedy but because copyright is the least efficient property system known.

Early speculation that AI models would act as infringement machines producing works like training material has proven unfounded, EFF says. Courts have already ruled that training generative AI models on copyrighted works is fair use because training serves a fundamentally different purpose, EFF notes. General-purpose AI models are not designed or used to create works aimed at any specific market, according to EFF. EFF argues that when unpermitted copying excludes fair use, the law cannot protect transformative uses that help everyone.

These models have created new insights and tools for medicine, science, journalism, art, and accessibility, EFF states. EFF notes that Be My AI depends on OpenAI’s GPT models to produce visual descriptions for people with blindness or low vision. Generative tools have sped up scientific research, and researchers have used them to develop treatments for conditions once considered incurable, according to EFF. Narrowing fair use would impede medical research, investigative reporting, and access to privacy protection tools, according to EFF.

Copyright exists to stimulate creativity and innovation for the greater public good, EFF states. EFF says copyright law serves to encourage new creation rather than shield authors who published first. Under the plaintiffs’ theory, no one could build on the work of predecessors, EFF argues. Copyright could block new books inspired by favorite novels or genres and academic literature reviews distilling prior research if readers might prefer the newer works, EFF says.

Some creators and publishers contend AI developers should get licenses, EFF reports. Creators often would not gain from these licensing deals, according to EFF. Large media and entertainment companies have taken advantage of underpaid working artists for many years, EFF states. Artists currently receive poor deals from record labels and publishers, and there is no reason to expect fairer treatment once those companies control AI training data, EFF argues.

EFF filed amicus briefs in two simultaneous U.S. copyright cases aimed at one novel legal theory. The Concord Music Group v. Anthropic case involves music publishers pressing claims against the AI lab about training data, and the In re Mosaic LLM Litigation case merges related claims into a class proceeding. The market dilution argument appears in both cases, claiming that outputs from AI systems compete with training works and that this alone is fair use’s market harm element.

EFF argues that this theory would grant rights holders veto power over competing expressive works and collapse factor four of fair use into a near-absolute bar. If courts accept market dilution, fair use would no longer function as a dependable compliance basis for training data ingestion, requiring explicit licenses for more content or risking direct infringement liability.

Wikimedia reports that The Times claims any use of its content, even for training, breaches copyright law, whereas OpenAI and Microsoft describe training on publicly available data as fair use. Wikimedia asks the court to review each use individually and points to the Supreme Court’s 2023 Warhol decision. Scraping a website is not itself a use, and only what a company does with the copies should count, Wikimedia argues.

The Wikimedia Foundation asked a federal judge in New York on Thursday for permission to file a friend-of-the-court brief in re OpenAI and announced the filing on Friday. Wikimedia argues training an AI model on copyrighted work is likely fair use but a chatbot copying websites to answer a user’s question is likely not. “We are taking a position in this case to defend the free and open human knowledge that Wikipedia depends on,” said Bernadette Meehan. The proposed Wikimedia brief supports neither party and states that both sides advance overbroad readings of fair use.

A chatbot using retrieval-augmented generation, a method in which a model pulls passages from sources to answer questions, copies text from sites such as Wikipedia or The New York Times when a user asks a question, feeds relevant passages to the model, and often reproduces much of the original in its answer. Wikimedia wrote that pre-training ChatGPT’s general LLM qualifies as fair use, but OpenAI’s reliance on copies in ChatGPT’s retrieval-augmented generation feature does not. Companies using retrieval-augmented generation should follow each publisher’s license terms, and the same applies to newer grounding techniques, the brief says.

For Wikipedia, the license is Creative Commons CC BY-SA 4.0, requiring credit to original authors and keeping adaptations under the same license. Wikimedia says that chatbots using Wikipedia content without observing those terms drive people away from contributing. Scraping strains Wikimedia’s servers and pushes the foundation to add technical safeguards that bad actors often get around. The foundation wants heavy users to follow public rate limits or pay for its Enterprise service.

This month, Wikimedia said that rogue OpenAI agents had made edits to its wikis without approval. The foundation says Wikipedia offers more than 67 million articles across 300 languages.

Nearly 400 local newspapers sued OpenAI and Microsoft in June. The Times and other outlets requested judicial sanctions against OpenAI in July, alleging that it concealed evidence. In September, the Justice Department told a New York court that AI training is fair use.

AI technologies will affect everyone and copyright law is poorly suited to handle many resulting consequences, EFF says. Technology rules should serve the public, not just companies able to litigate for decades or settle for billions, EFF argues. The outcome could reshape copyright protections for everyone, not just AI developers, because EFF argues the same legal theories would restrict researchers, journalists, accessibility tools, and ordinary internet activity.