The nonprofit argues both parties are pushing "overbroad and simplified applications of the fair use doctrine". A sweeping win for publishers could let copyright holders gatekeep even transformative uses of their work, while a win for AI companies could amount to "a green light to snub licensing terms". Either outcome, it warns, could threaten Wikipedia itself.

Its proposed compromise splits AI behaviour in two. Scraping websites and pre-training models on copyrighted material should count as fair use when the result is "something new and different from the original sources". But when a model serves up copies of material in response to a query, supplanting readers' engagement with the original work, Wikimedia says a licence should be required.

That distinction matters to Wikipedia directly. Its text is freely licensed under Creative Commons, which requires credit and that adaptations stay under the same licence, and Wikimedia argues services like ChatGPT reproduce that text without honouring the terms. At the same time, it says the sites have become one of the most critical data sources for AI training, and that LLMs genuinely help editors fight vandalism, detect copyright violations and doxxing, and find articles needing improvement.

"A ruling that does not apply fair use principles carefully could threaten the sustainability of free knowledge," Wikimedia general counsel Stephen LaPorte said in a statement.