Wikimedia Files Court Brief in AI Copyright Case

The Wikimedia Foundation has submitted a court brief in the copyright case brought by The New York Times against OpenAI and Microsoft. The foundation, which operates Wikipedia, does not fully support either side. According to the foundation, an overly strict and broad ruling could harm both the open internet and the development of new technologies.

The New York Times opposes AI companies freely using the material it publishes. OpenAI and Microsoft consider training AI models on publicly available data essential to technological development and a form of fair use. The foundation warns that a complete victory for publishers could restrict new ways of using content. A complete victory for AI companies, however, could lead to licensing rules being disregarded.

According to the foundation, the automated collection of information from websites should not, by itself, be considered copyright infringement. In some cases, the unlicensed use of copyrighted works during initial model training may also qualify as fair use. For this to apply, the output generated by the model must be new and different from the original source. If the model displays an article almost exactly as it appeared when answering a user’s question, that response then replaces the original material.

The foundation says the same approach also applies to RAG technology. When preparing a response, a RAG system does not rely solely on the model’s training data and also consults a separate, reliable knowledge base. If the system reproduces text from a source without attribution, the foundation does not consider this appropriate and stresses that a license is necessary. When Wikipedia material is used, the source must be cited, and modified content must also be shared under the same Creative Commons license.

According to the Wikimedia Foundation, Wikimedia sites have become one of the most important sources of data for AI and machine learning. However, chatbots are drawing users away from these sites and harming the work of the communities that maintain the content. On the other hand, these models can help editors combat vandalism. They can also be used to detect copyright violations, the disclosure of personal information, and articles that need to be updated.

A final ruling in the case is expected soon. This lawsuit is one of more than 12 similar AI copyright cases being considered by courts and could set an important precedent for future disputes. The foundation’s general counsel, Stephen LaPorte, believes that if fair use principles are not applied carefully, the sustainability of free knowledge and open-source technologies could be harmed.