Wikimedia Foundation argues AI training is fair use but RAG responses are not in OpenAI lawsuit

The Wikimedia Foundation has intervened in the copyright lawsuit between The New York Times and OpenAI, filing an amicus brief that distinguishes between two uses of copyrighted content: training AI models and generating chatbot responses. The foundation argues that training on such content is likely fair use, but using it in retrieval-augmented generation (RAG) responses may not be. The outcome could shape how AI companies use publicly available data.

Tarun Mottlia

Editor, Lazyfounder

Published Nov 1, 20263 min read

Wikimedia Foundation argues AI training is fair use but RAG responses are not in OpenAI lawsuit — Image: Sasun Bughdaryan on Unsplash via source

Wikimedia’s position in the OpenAI lawsuit

The Wikimedia Foundation has filed an amicus brief in the copyright infringement case between The New York Times and OpenAI, urging the court to evaluate the use of copyrighted material in AI training and chatbot responses separately. According to The Next Web, the foundation argues that training AI models on copyrighted content is likely fair use, but using such content in retrieval-augmented generation (RAG) responses—where a chatbot reproduces portions of the original text—is not.

The brief references the U.S. Supreme Court’s 2023 ruling in Andy Warhol Foundation v. Goldsmith, which emphasized that each use of copyrighted material must be assessed individually. Wikimedia contends that scraping a website does not itself constitute a "use" under copyright law; only how the copied material is later used matters.

Wikipedia’s licensing and AI’s impact

Wikipedia’s content is licensed under Creative Commons CC BY-SA 4.0, which requires attribution to original authors and mandates that adaptations remain under the same license. Wikimedia’s general counsel, Stephen LaPorte, told The Next Web that when AI chatbots use Wikipedia content without adhering to these terms, it discourages contributions and undermines the platform’s open knowledge model.

The foundation also highlighted the technical strain caused by scraping. Wikimedia has implemented safeguards to manage server load and encourages heavy users to comply with rate limits or subscribe to its Enterprise service, which offers structured access for commercial users.

Broader implications for AI and publishing

The lawsuit, pending before Judge Sidney H. Stein in the U.S. District Court for the Southern District of New York, consolidates 12 cases against OpenAI, including The New York Times’ 2023 suit. The Times alleges that OpenAI’s use of its content—both for training and in chatbot responses—violates copyright law. OpenAI and Microsoft, on the other hand, maintain that training on publicly available data qualifies as fair use.

Wikimedia’s brief aligns partially with the U.S. Justice Department, which in September argued that AI training is fair use. However, the foundation’s stance on RAG responses introduces a nuanced distinction that could influence the court’s ruling. Nearly 400 local newspapers sued OpenAI and Microsoft in June, signaling broader industry concerns about AI’s use of copyrighted material.

What this means

Lazyfounder analysis — our interpretation, not reported fact.

This filing is a strategic move for Wikimedia, which relies on open knowledge while facing server strain from scraping. For founders and operators in AI, the case highlights the growing legal risks around RAG systems, which often reproduce chunks of original text. If the court sides with Wikimedia’s distinction, AI companies may need to rethink how they deploy retrieval-augmented responses—particularly for content under restrictive licenses like Creative Commons. This could also push more businesses toward structured, paid access models like Wikimedia’s Enterprise service, or toward proprietary datasets that offer clearer terms of use.

Sources

The Next Web · 2026-10-30
Wikimedia asks court to rule AI training fair use, but not chatbot copying

This story is an original summary drafted with AI by Lazyfounder from the reporting listed above and checked by automated validation. Facts are attributed to their original publishers; sections marked as analysis are Lazyfounder's. Where a source is in another language, facts were machine-translated and quotations are reported, not reproduced. Read the original coverage via the links, and see our AI policy and corrections policy.

About the author