The Wikimedia Foundation has asked courts weighing AI copyright disputes to draw a distinction between training a model on copyrighted material and using a chatbot to reproduce that material. Its position aims to preserve AI tools that can help Wikipedia editors while protecting the people and communities whose work supplies knowledge to the open web.
Wikimedia argues against an all-or-nothing ruling
The nonprofit behind Wikipedia filed an amicus brief in lawsuits brought by publishers, including The New York Times, against OpenAI and Microsoft. It does not take either side. Instead, Wikimedia warns that a broad ruling in favor of either publishers or AI companies could have consequences beyond the parties in the cases.
If copyright holders win a sweeping right to control uses of their work, Wikimedia worries that they could block innovative and transformative applications. If AI companies prevail without meaningful limits, the result could encourage them to ignore licensing terms and weaken the connection between authors and audiences.
The Foundation says courts should apply fair use with care. The question, in its view, is not simply whether an AI system encountered copyrighted work, but what the system does with that material and whether its output substitutes for the original.
Training can support tools for Wikipedia editors
Wikimedia says its websites have become important sources of material for AI and machine learning training. At the same time, the Foundation says AI services can draw visitors away from the sites and communities that create and maintain that knowledge.
Those systems also have practical uses for Wikipedia. Large language models can help editors identify vandalism, spot possible copyright violations or doxxing, and find articles that need improvement. A ruling that broadly prevents training could therefore limit tools that support the work of volunteers and the availability of free knowledge.
That concern informs Wikimedia’s proposed distinction: scraping websites and using copyrighted works to pre-train large language models without a license should count as fair use when the models produce something new and different from their sources.
Reproducing source material raises a different issue
The Foundation draws a line at chatbot responses that return copies of source material to users. If a system supplies text in a way that replaces readers’ engagement with the original work, Wikimedia argues, that output may not transform the material enough to qualify as fair use. In such cases, it says, a license should be required.
This distinction focuses on the effect of the output. A model’s training may serve a different purpose from the underlying articles, while a response that reproduces those articles can act as a substitute for reading them. Wikimedia says AI companies should respect licensing requirements when they use information in that substitutive way.
For Wikipedia content, the Foundation points to the terms of its Creative Commons license. Those terms require crediting the original source and require remixes or adaptations to remain under the same license. Wikimedia says services such as ChatGPT reproduce text that is freely licensed under Creative Commons but do not follow those conditions.
The dispute matters to the open web
The New York Times alleges that OpenAI infringed its copyright by scraping its website to build training datasets. OpenAI and Microsoft argue that using material to train large language models is fair use. OpenAI has described regurgitating work as “a rare bug.” Microsoft has said expert analysis found that Copilot rarely reproduces full sentences from news content used to ground the model, and that this does not undermine the transformative purpose of training.
Wikimedia’s filing raises a broader concern about how the ruling could shape access to knowledge. The Foundation says Wikipedia depends on open, accessible human knowledge and a collaborative process. Its position is that AI can expand access to information, but the people and communities that make reliable knowledge possible must be respected.
The outcome could affect both sides of that relationship: whether AI developers can use copyrighted works to build tools, and whether their products can return those works to users without permission. For Wikimedia, preserving free knowledge means treating those activities separately and applying copyright rules with attention to what AI systems actually provide.