Wikimedia warns bad AI copyright ruling could threaten the open web

The Wikimedia Foundation has staked out a middle ground in a group of landmark AI copyright lawsuits. The organization is urging courts to legally protect large language models trained on copyrighted material, while pushing back against chatbots regurgitating that exact work to end users.

• Wikimedia argues AI training is fair use, but regurgitating original articles should not be;
• An overly black-and-white ruling could deprive editors of AI tools or endanger the open web;
• The nonprofit filed an amicus brief weighing in on lawsuits against OpenAI and Microsoft.

Wikimedia's nuanced legal stance

In an amicus brief filed Friday, the nonprofit behind Wikipedia weighed in on copyright lawsuits involving publishers like The New York Times, OpenAI, and Microsoft. Wikimedia is concerned that current legal positions push for overly broad and simplified applications of the fair use doctrine. A nuance-free victory for publishers could let copyright holders gatekeep innovative and transformative uses of their work.

Conversely, an unconditional victory for AI companies would give a green light to snub licensing terms and fragment the flow of knowledge. Wikipedia websites have become critical data sources for machine learning training, yet these systems often siphon away visitors, undermining communities that maintain the sites and failing to respect Creative Commons terms.

The dual role of AI for Wikipedia

At the same time, large language models are genuinely useful to Wikipedia editors. They help fight vandalism, detect copyright violations, prevent doxxing, and flag articles that need improvement. Wikimedia general counsel Stephen LaPorte noted that a ruling failing to apply fair use carefully could threaten the sustainability of free knowledge and stifle open-source technologies.

The foundation emphasizes that Wikipedia is created through a collaborative process requiring a strong foundation of open human knowledge. While OpenAI claims text regurgitation is a rare bug and Microsoft points to the transformative purpose of model training, Wikimedia highlights the necessity of maintaining a clear boundary.

A proposed compromise for the information ecosystem

Wikimedia proposes a clear compromise: web scraping and pre-training LLMs on copyrighted works without a license should constitute fair use if they provide something new and different. However, if models pull copies of material in response to user queries and supply them directly, supplanting original readership, that output must require a license.

CEO Bernadette Meehan stated that defending free and open human knowledge is a shared responsibility. For developers, researchers, and tech communities worldwide, this debate establishes vital precedents on balancing technological advancement with respect for human contributors and reliable knowledge ecosystems.