Wikimedia urges court to separate AI training from chatbots reproducing content

The Wikimedia Foundation, the nonprofit that operates Wikipedia, filed an amicus brief on Friday in a group of lawsuits in which publishers, including The New York Times, accuse AI companies OpenAI and Microsoft of copyright infringement. Wikimedia sides with neither party, but warns that an overly simplified ruling could harm both Wikipedia and the open web.

Concerns about both outcomes

The organization says the parties' positions push for overbroad and simplified applications of the fair use doctrine. If publishers win outright, copyright holders could gatekeep even innovative and transformative uses of their work. If AI companies win, it could be seen as a green light to ignore licensing terms and fragment the flow of knowledge between human authors and audiences.

Wikimedia says its sites have become one of the most critical data sources for AI and machine learning training. At the same time, these systems draw visitors away and undermine the communities that maintain the sites. Wikimedia also argues that services such as ChatGPT reproduce text freely licensed under Creative Commons without following the terms of that license. Yet large language models are also useful to Wikipedia, helping editors fight vandalism, detect copyright violations and doxxing, and find articles that need improvement.

The proposed compromise

Wikimedia argues that scraping websites should not count as copyright infringement, and that pre-training language models on copyrighted works without a license should be fair use when the result is something new and different from the original sources. But when models pull copies of material in response to user queries and deliver it while supplanting readers' engagement with the originals, a license should be required. For Wikipedia, that would mean crediting the original source and keeping any remix or adaptation under the same Creative Commons license.

The Times alleges that OpenAI infringed its copyright by scraping its website to build training datasets. OpenAI and Microsoft contend that using language models is fair use. OpenAI has called regurgitation a rare bug, while Microsoft said in a September filing that expert analysis found Copilot rarely reproduces full sentences from news content.

Wikimedia CEO Bernadette Meehan said that expanding access to knowledge through AI is a powerful goal, but it must respect the people and communities that make reliable knowledge possible.