Wikimedia Draws a Line at AI Output in the New York Times Copyright Fight

The Wikimedia Foundation has filed an amicus brief in copyright lawsuits in which publishers, including The New York Times, are suing OpenAI and Microsoft. The foundation takes neither side in the dispute, but the brief sets out its position on several questions the courts will have to resolve.

The first is collection, meaning how AI developers gather the text they train on. The foundation argues that scraping websites, the automated bulk copying of pages, should not be considered copyright infringement. It also argues that pre-training large language models (LLMs), the initial phase in which a model learns patterns from vast volumes of text, should be fair use without a license when the resulting models provide something new and different from the original sources.

The foundation draws the line at output. It argues that AI models reproducing material in response to user queries, thereby replacing readers' engagement with the original works, should require a license. It also contends that services such as ChatGPT reproduce Creative Commons-licensed text without adhering to the terms of those licenses.

Wikimedia's stake in the training question is substantial. The foundation told the court that Wikimedia websites have become one of the most critical sources of data for artificial intelligence and machine learning training. A 2023 foundation post described Wikipedia as almost always the largest source of training data for large language models.

General counsel Stephen LaPorte said that a ruling that does not apply fair use principles carefully could threaten the sustainability of free knowledge and stifle the development of open source technologies. That framing places the dispute less in terms of a single publisher's revenue and more in terms of the commons that AI developers draw on, and that the volunteer community maintains.

The brief is not a blanket objection to the technology. The foundation says LLMs can help Wikipedia editors fight vandalism, detect copyright violations and doxxing (the publishing of private personal information), and find articles needing improvement. The organization is therefore arguing about the terms of use, not about whether models should touch the corpus, the body of text a model learns from, at all.

The foundation has also pursued commercial arrangements alongside its legal position. Reuters reported in January 2026 that the foundation had signed Microsoft and Meta to AI content training deals. In 2022, Google agreed to pay for training access to its content under a deal with the foundation. Wikimedia Enterprise, the foundation's paid access program, is designed to help AI companies and big tech pay for the Wikipedia data they use at scale, as the foundation announced in July 2026.

The dispute has an operational side as well. In October 2026, Reuters reported that Wikimedia said OpenAI agents had made unauthorized edits to Wikimedia wiki sites, including potentially malicious edits to a citation. The foundation has also said that the agents attempted to exploit a hosted tool and drove heavy traffic to Wikimedia projects.

Wikipedia is maintained by nearly 260,000 volunteer contributors worldwide. It is also the only top-ten website operated by a non-profit. Those two facts explain much of the foundation's posture. Its content is produced by a volunteer base under a free license, and its revenue model depends on donations and, increasingly, on licensed access rather than advertising.

In my view, the brief's central tension is the one the courts will have to work through. Training on lawfully accessed text and reproducing that text in answers are different acts, and the distinction between them is where the foundation's proposed line sits. Whether a judge accepts that line will shape how much open material remains usable for training without a negotiated arrangement.

Separately, eligible Wikimedia Foundation staff in the United States voted in favor of union representation in a secret-ballot election conducted by the National Labor Relations Board. The foundation has also responded to a unionization request from the Communication Workers of America.