In a recent filing related to the New York Times lawsuit against OpenAI, the Wikimedia Foundation warned that an adverse court decision could endanger the open web. The nonprofit, which operates Wikipedia and its sister sites, asked judges to draw a distinction between using copyrighted material to train models and the models’ later verbatim output.
Wikimedia argues its sites have become among the most vital sources of data for artificial-intelligence and machine-learning training. At the same time, the organization notes that AI services draw traffic away from its pages, weakening the volunteer communities that maintain the content. It further claims that platforms such as ChatGPT reproduce text released under a Creative Commons licence without honoring the licence’s attribution and share-alike conditions.
The foundation maintains that using copyrighted works to pre-train large language models should fall under fair use, provided the training creates something novel that does not simply copy the original. By contrast, when an AI system returns a passage that mirrors a Wikipedia article, the output replaces the reader’s need to visit the source, which the filing says warrants a licence.
Wikimedia’s proposed middle ground calls for courts to treat web-scraping as non-infringing, while recognizing that unlicensed use of copyrighted material for model training may be permissible under fair use if it yields a transformative result. However, the organization insists that any AI-generated text that directly substitutes the original work must respect the underlying Creative Commons terms, including attribution and the requirement that derivative works remain under the same licence.
The filing also highlights potential benefits of large language models for Wikipedia’s editorial workflow. It cites uses such as automatically spotting vandalism, flagging possible copyright breaches, and identifying articles that need expansion. By securing a legal framework that permits training while obligating proper licensing of reproduced content, Wikimedia hopes to preserve both the free-knowledge ecosystem and the tools that can enhance it.