The Wikimedia Foundation, which runs Wikipedia, is staking out a middle ground in a set of AI copyright lawsuits without siding with either party. In an amicus brief filed on Friday, it asks courts to keep two things apart: training a model on copyrighted texts, which it argues could fall under fair use, and letting a chatbot hand those same texts back to users, which it says should not.
The most prominent case is The New York Times against OpenAI and Microsoft, alongside other publishers. The paper argues OpenAI infringed its copyright by scraping its website to build training datasets. OpenAI and Microsoft respond that using large language models is fair use, and that reproducing text is a rare glitch. Microsoft, in a September filing, cites expert analysis finding that Copilot seldom reproduces full sentences from the news content used to ground its answers.
Wikimedia fears that a clean win for publishers could let rights holders block even innovative and transformative uses of their work. A clean win for AI companies, on the other hand, could let firms ignore licences and split the flow of knowledge between authors and readers. For Wikipedia the stakes are concrete: its pages have become one of the most important sources for machine learning training, and the same systems draw visitors away from the sites that maintain that content.
The foundation also argues that services like ChatGPT reproduce text released under Creative Commons without respecting its terms. Yet it acknowledges that AI models help Wikipedia's volunteers, who use them to spot vandalism, copyright violations and doxxing, and to find articles that need work.
Its proposed compromise is specific. Scraping websites should not count as copyright infringement. Using protected works to pre-train models without a licence should qualify as fair use when the output offers something new and different from the sources. But when a model pulls copies of material to answer a user and thereby replaces the reader's engagement with the original, a licence should be required, because the output then does not transform the work enough to count as fair use. In those cases, companies should honour licence conditions, which for Wikipedia means attribution and keeping derivatives under the same licence.
CEO Bernadette Meehan said the organisation is defending the free human knowledge Wikipedia depends on. Expanding access to knowledge through AI is a worthy goal, she said, but it has to respect the people and communities who make reliable information possible, and striking that balance is a shared responsibility across the information ecosystem.
Earlier this month the foundation had already reported on its wikis "rogue" OpenAI agents that crawled millions of pages and attempted unauthorised edits without compromising any systems, an episode that makes the legal question more pressing.
No comments yet. Be the first!