Wikipedia’s operator says a bad AI copyright ruling could threaten the open web

Wikipedia’s Operator Warns of an Existential AI Copyright Threat

The Wikimedia Foundation, the nonprofit behind Wikipedia, has filed an amicus brief in the high‑profile lawsuits involving The New York Times, OpenAI, and Microsoft. While the foundation does not side with either plaintiffs or defendants, it stresses that a simplistic court ruling on fair use could have far‑reaching consequences for the open web and for Wikipedia’s sustainability.

Why Wikipedia Matters to AI

Wikipedia is one of the most heavily used data sources for training large language models (LLMs). Its openly licensed content—primarily under Creative Commons Attribution‑ShareAlike—provides a rich, structured knowledge base that AI developers rely on to teach models how to understand and generate human language. At the same time, AI tools such as ChatGPT can pull directly from that same content, sometimes reproducing passages verbatim. The foundation argues that:

• Training on Wikipedia is generally fair use because it creates a transformative product that offers new capabilities beyond the original articles.
• Regurgitating Wikipedia text to users is not; it substitutes the original work without adding sufficient transformation, violating the spirit of the CC‑BY‑SA license.

The Two‑Sided Risk

The brief outlines a “double‑edged sword” scenario:

1. If courts favor publishers and deem any use of copyrighted material for LLM training as infringement, AI developers may be forced to obtain licenses for every source they scrape. This could make it prohibitively expensive to train models, stifling innovation and limiting the availability of free AI tools that benefit Wikipedia editors—such as vandalism detection, citation suggestions, and article‑improvement recommendations.
2. If courts side with AI companies, the decision could be interpreted as a green light for unlicensed data harvesting. That would allow AI services to continue reproducing Wikipedia content without attribution, undermining the collaborative model that relies on volunteers and potentially eroding traffic to the encyclopedia.

Both outcomes threaten the open‑knowledge ecosystem that Wikipedia helps sustain.

A Proposed Middle Ground

Wikimedia’s position is nuanced. It suggests:

• Scraping publicly available web pages for model training should be permissible under fair use when the output is transformative and does not replace the original work.
• Providing verbatim excerpts in response to user queries should require a license and proper attribution, respecting the CC‑BY‑SA terms.

The foundation also calls for AI developers to credit sources and to keep any derivative works under the same open license, preserving the “share‑alike” principle that fuels collaborative improvement.

What This Means for the Future of the Open Web

A ruling that ignores these distinctions could reshape how knowledge is shared online:

• Content creators may become gatekeepers, demanding licensing fees that limit the free flow of information.
• Open‑source and community‑driven projects could lose the data pipelines that make modern AI possible, reducing the tools available to educators, journalists, and the public.
• Wikipedia’s own community might see a decline in traffic and contributions if AI systems siphon away readers.

The Wikimedia Foundation urges courts to apply fair‑use doctrine carefully, recognizing the collaborative nature of Wikipedia and the broader public interest in maintaining an open, accessible web.

---

Bottom line: A balanced legal approach is essential. Overly broad copyright restrictions could choke the data that powers AI, while unchecked AI use could erode the very open‑knowledge foundations Wikipedia relies on. The outcome of these cases will likely set the tone for how the internet’s most valuable public resource coexists with the next generation of AI.