Wikimedia Asks Court to Split Fair Use in OpenAI Case

The Wikimedia Foundation filed an amicus brief on Friday in the consolidated copyright litigation against OpenAI in the Southern District of New York, arguing that the court should treat AI model training and chatbot retrieval as two different legal questions. The nonprofit behind Wikipedia is not backing either side in the case, which combines twelve lawsuits including The New York Times Co. v. Microsoft and OpenAI.

What Wikimedia is asking for

The brief splits fair use into two buckets. Using copyrighted material to pre-train an LLM should count as fair use, the Foundation argues, as long as the resulting model is "something new and different from the original sources." In that case no licensing terms would apply.

Retrieval-augmented generation is the opposite case. When a system fetches a news article or other work from the web and serves that content back to users through a chatbot, Wikimedia says fair use should not apply and companies "should respect each content platform or publisher's licensing requirements." For Wikipedia that means CC BY-SA 4.0: credit to original creators, and any remix or adaptation staying under the same license.

"Wikimedia is concerned that the parties' positions in this case push for overbroad and simplified applications of the fair use doctrine," the filing states. A blanket win for publishers could let rights holders "gatekeep" transformative uses; a blanket win for AI companies could be "a green light to snub licensing terms and fragment the flow of knowledge."

Why it matters

The ruling will not be a footnote. The case is before Judge Sidney Stein in the most influential court for AI copyright, and only a handful of precedents exist anywhere, so whatever is decided here will shape how other courts read fair use for years. The Times argues that training on its content is infringement outright, while OpenAI and Microsoft say LLM training and deployment fall inside fair use.

For Wikipedia itself the stakes are direct. Wikimedia calls its projects "one of the most critical sources of data for all artificial intelligence and machine learning training," while noting those same systems siphon away readers who sustain the volunteer communities producing the content. It also uses LLMs internally to fight vandalism, flag copyright violations and doxxing, and find weak articles.

"Expanding access to knowledge through AI is a powerful goal, but it must respect the people, contributors, and communities that make reliable knowledge possible," CEO Bernadette Meehan said.

Builder impact

The practical split matters more than the legal theory. If the court follows Wikimedia's framing, training data stays largely unlicensed territory but the retrieval layer becomes a licensing obligation. Teams running RAG over news, documentation or community content should treat source licensing as an architecture constraint now, not a compliance item after launch: track per-source license terms, keep attribution attached to retrieved chunks, and know which sources must be excluded from grounding pipelines.