The nonprofit behind Wikipedia wants courts to protect AI training on copyrighted work while drawing a hard line at chatbots that regurgitate it
The Wikimedia Foundation has entered the biggest AI copyright fight in the US, and it is not picking a team.
In an amicus brief filed Friday, the nonprofit behind Wikipedia asked courts to take a middle path. Training large language models on copyrighted material should get legal protection, it argues. Chatbots spitting that material back out to users should not.
That position puts Wikimedia in an unusual spot. It is effectively telling both The New York Times and OpenAI that they are each partly right.
What Wikimedia is asking for
The filing lands in In Re: OpenAI, Inc. Copyright Infringement Litigation, a consolidated case in the Southern District of New York under case number 1:25-md-03143. The suits pit publishers, The New York Times among them, against OpenAI and Microsoft.
Wikimedia’s brief splits the AI pipeline into two distinct stages. The first is training, where a model ingests huge volumes of text to learn patterns in language. The second is output, where a chatbot answers a user’s question.
On training, the Foundation argues that feeding copyrighted works into an LLM may qualify as fair use. Fair use is the legal doctrine that lets people use protected material without permission in certain circumstances, such as commentary, research or transformation into something new.
On output, Wikimedia draws a firmer line. A chatbot that regurgitates protected work verbatim to users is a different animal, and the Foundation does not want that behavior swept under the same protection.
Why an all-or-nothing ruling worries Wikipedia
The Foundation’s core concern is that the two sides are pushing the court toward overly broad conclusions.
If the court rules that training itself is infringement, Wikipedians could lose access to useful AI-based tools. Volunteer editors increasingly lean on software to help maintain a sprawling encyclopedia, and a sweeping ban on training could choke off development of those helpers.
If the court instead blesses AI outputs wholesale, including verbatim reproduction, Wikimedia sees a different danger. A ruling that favors outputs could discourage people from sharing knowledge online at all.
Projects like Wikipedia depend on people choosing to publish freely. If anything posted publicly can be absorbed and served back by a chatbot without limits, the incentive to contribute starts to look thinner.
A crowded courtroom
Wikimedia is far from the only outside voice in this case. Related amicus briefs from publisher groups and unions landed around the same time.
The Association of American Publishers has filed briefs opposing broad fair-use protections for AI developers.
The US government has also weighed in. In September 2026, it filed a statement of interest supporting fair use for AI training.
The underlying dispute is not new. The allegations stretch back to 2023, when publishers began accusing AI developers of using their copyrighted articles without authorization to build their models.
Since then, the litigation has grown into a consolidated proceeding. Whatever the court decides on the training-versus-output question will likely echo well beyond OpenAI and Microsoft.
