The Wikimedia Foundation is asking courts to protect the training of large language models on copyrighted material while refusing to shield chatbots that reproduce that material for users, according to an amicus brief the nonprofit filed Friday in the copyright lawsuits brought by publishers including The New York Times against OpenAI and Microsoft. The foundation, which operates Wikipedia, said it is not backing either side in the case.
In its filing, the Wikimedia Foundation warned that the parties' positions push for overly broad and simplified applications of fair use doctrine to AI. It said a sweeping win for publishers could allow copyright holders to gatekeep even innovative and transformative uses of their work, while a win for AI companies could signal that licensing terms can be ignored and that the flow of knowledge between human authors and audiences can be fragmented.
The foundation said its own sites have become one of the most important data sources for AI and machine learning training, but that these systems pull away visitors and undermine the communities that maintain those sites. It also argued that services such as ChatGPT reproduce text licensed freely under Creative Commons without following the terms of that license.
At the same time, the filing said large language models are useful to Wikipedia, including for fighting vandalism, detecting copyright violations and doxxing, and finding articles that need improvement. Stephen LaPorte, the foundation's general counsel, said in a statement that a ruling that does not apply fair use principles carefully could threaten the sustainability of free knowledge and stifle open source development, affecting access to information in both cases. He said courts need to understand that Wikipedia depends on a strong foundation of open and accessible human knowledge.
The Times has alleged that OpenAI infringed its copyright by scraping its website to build training datasets, and it seeks a broader reading of fair use as applied to large language models. OpenAI and Microsoft both argue that training on such material constitutes fair use; OpenAI has described regurgitating work as a rare bug, and Microsoft said in a September filing that expert analysis found Copilot rarely reproduces full sentences from news content used to ground the model.
The Wikimedia Foundation proposed a middle path: scraping websites should not count as infringement, and using copyrighted works to pre-train models without a license should qualify as fair use when the output offers something new and different from the originals. But when a model returns copies of material in response to user queries and replaces readers' engagement with the original works, the foundation said, a license should be required, with credit given under Creative Commons terms that also require remixes and adaptations to stay under the same license.
CEO Bernadette Meehan said in a statement that the foundation is taking a position to defend the free and open human knowledge Wikipedia depends on. She said expanding access to knowledge through AI is a powerful goal but must respect the people, contributors, and communities that make reliable knowledge possible, calling that balance a mutual benefit and a shared responsibility for the information ecosystem.
More AI news from TechManNews.
