Wikimedia asks court to separate AI training from copied answers
Wikimedia Foundation has asked a US court to distinguish AI model training from chatbot answers that reproduce source material. The operator of Wikipedia supports fair use for training that creates something different, while calling for licensing terms to govern copied answers. Its intervention highlights Wikipedia’s reliance on both useful AI tools and a sustainable supply of human-written sources.
Artificial Intelligence··Evening
Court filing separates training from content delivery
Wikimedia Foundation, the nonprofit operating Wikipedia, has intervened in US copyright litigation involving OpenAI and Microsoft with a proposed distinction between training a model and copying material into its answers. The foundation announced its amicus brief on October 30. The consolidated proceeding in the Southern District of New York combines 12 lawsuits, including the case The New York Times brought in 2023. Wikimedia takes neither side outright: it asks the court to assess different uses of source material separately.[1], [2]
Its proposal would protect training under fair use when the result is something new and different from the underlying works. Wikimedia says an excessively restrictive interpretation could allow copyright owners to block innovative uses, including tools benefiting open-source projects. At the same time, it warns that blanket permission for AI companies could undermine the publishers and human authors on which open knowledge relies.[1], [2]
Copied answers would retain license obligations
The foundation draws a different line for retrieval-augmented generation, where a system finds web material and delivers copied content through a chatbot. When that answer substitutes for engagement with the original work, Wikimedia argues that publishers’ licensing requirements should apply. Wikipedia uses the Creative Commons CC BY-SA 4.0 license: users must credit original creators and distribute adaptations under the same license. The foundation also calls for sustainable scraping, through public application programming interface rate limits or its Enterprise API services.[1], [2]
Wikimedia’s general counsel Stephen LaPorte says a careless application of fair use could threaten sustainable free knowledge and the development of open-source technologies. Chief executive Bernadette Meehan says expanding access through AI must respect the people, contributors and communities creating reliable knowledge. Those positions accompany the foundation’s request to preserve the connection between human authors and their audiences.[2]
Wikipedia uses AI while depending on human sources
Wikimedia says language models can help Wikipedia editors find vandalism, copyright problems, disclosures of personal information and articles needing improvement. Its encyclopedia contains more than 67 million articles in 300 languages, with nearly 250,000 volunteers contributing each month. Those contributors depend on secondary sources such as researched news reporting. OpenAI describes verbatim reproduction as a rare software fault; Microsoft says analysis of Copilot found full sentences from grounding news were rarely reproduced. These remain the parties’ positions in ongoing litigation, alongside Wikimedia’s requested distinction.[1], [2]
