Wikimedia asks court to separate AI training from real-time chatbot copying

The split legal strategy

The Wikimedia Foundation is asking a federal judge to split a complex copyright question into two distinct legal issues. They are doing this through an amicus, or friend-of-the-court, brief in a lawsuit involving OpenAI and The New York Times. The foundation wants the court to recognize that different ways of using data require different legal standards.

Currently, the case involves 12 lawsuits against OpenAI, including a 2023 suit from the Times. Judge Sidney H. Stein is tasked with deciding if using copyrighted content to build large language models qualifies as fair use. Wikimedia argues that both the plaintiffs and the defendants are pushing overbroad and simplified readings of the law.

The foundation's goal is to defend the free and open human knowledge that Wikipedia relies on. They believe the current legal debate ignores the nuances of how AI actually interacts with information. By splitting the question, they hope to protect the ability to build models while still protecting the rights of original creators.

Training vs. Retrieval

Wikimedia makes a sharp distinction between the 'pre-training' phase of an AI and the 'retrieval-augmented generation' (RAG) phase. Training involves feeding a model vast amounts of data to help it learn general patterns and information. The foundation argues this is likely fair use because the resulting model is a new product that does not replace the original works.

RAG is a different process where a chatbot copies text from websites, like Wikipedia or the Times, to answer a specific user question. In this scenario, the model feeds relevant passages into its system and often reproduces much of the original text in its response. Wikimedia argues that this specific use is not fair use because it directly mirrors the source material.

The foundation suggests that companies using RAG should be required to follow each publisher's specific license terms. They argue that the law should focus on what a company does with the copies rather than the act of scraping itself. This distinction relies on a 2023 Supreme Court ruling regarding the artistic work of Andy Warhol.

Impact on the ecosystem

For Wikipedia, these legal battles have practical consequences for its community of volunteers. When chatbots use content without following the Creative Commons CC BY-SA 4.0 license, it can steer people away from contributing to the site. The license requires that authors receive credit and that any adaptations remain under the same open terms.

The foundation also points out that heavy scraping puts a significant strain on its servers. This technical burden has forced Wikimedia to add safeguards that some bad actors frequently find ways to bypass. They want heavy users to either follow public rate limits or pay for a dedicated Enterprise service.

The foundation recently reported that rogue OpenAI agents were editing its wikis without any prior approval. These incidents highlight the friction between automated AI agents and the human-led systems that organize global knowledge. Wikimedia wants to ensure that AI development respects the communities that make reliable information possible.

The broader copyright fight

This case is part of a massive legal push by publishers to limit how AI companies use their content. In June, nearly 400 local newspapers filed a lawsuit against OpenAI and Microsoft. Other outlets have also asked the court to sanction OpenAI for allegedly hiding evidence during the litigation process.

The legal landscape is shifting as different government bodies weigh in on the issue. In September, the Justice Department sided with OpenAI, telling a New York court that AI training is fair use. This creates a complex environment where different courts may reach different conclusions based on how they interpret the facts.

Wikimedia's counsel argues that the conversation around copyright in the age of AI must include more than just for-profit organizations. They believe it must involve the entire online information ecosystem. The foundation wants a legal framework that balances the goal of expanding access to knowledge with the rights of the people who provide it.