The Wikimedia Foundation filed a neutral amicus brief arguing courts should separate AI training from content retrieval when weighing fair use
The Wikimedia Foundation has entered the copyright fight over OpenAI’s AI models. It isn’t taking either side.
On October 30, 2026, the nonprofit behind Wikipedia filed an amicus brief in In re: OpenAI, Inc. Copyright Infringement Litigation. Its warning is that a ruling drawn too broadly in either direction could damage the open structure of the web.
A brief that backs neither party
An amicus brief is a filing from an outside party that isn’t directly involved in a case. The term translates to “friend of the court.” These briefs let interested groups explain how a decision might ripple beyond the courtroom.
Wikimedia’s filing is characterized as supporting neither party. The foundation is effectively saying both sides have a point, and the judge should be careful about which point wins.
The core of the argument is a distinction between two ways AI systems use copyrighted material.
The first is pre-training. This is the foundational stage where a model learns patterns from a huge body of text. Wikimedia argues this use typically qualifies as fair use.
The second is retrieval-augmented generation, usually shortened to RAG. RAG-style systems fetch specific content and present it directly in their answers. Wikimedia’s position is that this kind of direct replication and presentation should follow licensing requirements.
Why Wikipedia cares so much
Wikipedia comprises over 67 million articles in more than 300 languages. It is one of the largest freely available bodies of human knowledge ever assembled.
The research summary of the filing lays out the two failure modes. A decision that’s too restrictive could hamper technological tools that benefit the volunteers who improve Wikipedia. A decision that’s too permissive could push publishers to limit access to their content.
Stephen LaPorte, Wikimedia’s general counsel, stated that incorrect rulings could threaten the sustainability of free knowledge.
CEO Bernadette Meehan highlighted the importance of defending free and open human knowledge, which she described as vital to Wikipedia.
What this means for AI companies and publishers
For AI developers, the argument is a mixed bag. The pre-training half of the brief is friendly territory, since it treats foundational learning as typically fair use. The RAG half is less comfortable, because retrieval features are increasingly how AI products deliver current, sourced answers. If courts adopt Wikimedia’s line, those features could come with licensing costs attached.
For publishers, the logic flips. They get no blanket win on training data under this framework. But they would gain leverage over products that reproduce and present their work directly.
