The Wikimedia Foundation, which hosts Wikipedia, has asked the federal court overseeing the OpenAI copyright cases to treat ChatGPT’s web retrieval differently from model training. In a proposed amicus brief filed on 29 October, it argues that training a general-purpose model is likely fair use, while the retrieval step is likely not fair use.
The brief puts the line this way: “It is likely not fair use when an LLM-based service uses copies of original works for their intended purposes—to provide relevant information, perspectives, expressions, and context—while supplanting readers’ engagement with those original works.”
Business Pill · Data Flywheel Explained
An evergreen Business Engineer explainer on data flywheels. It teaches the general idea only and does not discuss this case.
The key insight: Wikimedia’s proposed brief, filed 29 October, asks the court to split the fair-use question: training a general-purpose model likely fair, ChatGPT-style retrieval at query time likely not fair.
The filing is a letter-motion to Judge Sidney H. Stein of the Southern District of New York, in In re: OpenAI, Inc. Copyright Infringement Litigation. Wikimedia’s lawyers at Lex Lumina ask for leave to file the brief, which is titled as being “in Support of Neither Party”.
Wikimedia says the case gathers 12 combined lawsuits against OpenAI, the best known being The New York Times Co. v Microsoft and OpenAI. The Times filed in 2023.
The letter says retrieval-augmented generation, or RAG, harms sites like Wikipedia by “burdening and displacing access to its hosted content.” It adds that RAG’s significance “is not explored deeply by either party.”
How Wikimedia’s proposed amicus brief, filed 29 October 2026, splits the fair-use question by stage: scraping is not a use, pre-training is likely fair, RAG retrieval is likely not fair. Source: Wikimedia Foundation filing and statement.
Training Versus Retrieval
The brief asks the court to judge each use of a copied work on its own. On training, it says “the use of copyrighted works to train a general purpose LLM is likely to be fair under the four-step analysis.”
Retrieval is the step where a chatbot searches the web at the moment a user asks a question. The brief describes the service as one that copies text from sites “like Wikipedia or The New York Times”, pastes the most relevant chunks into the model’s context window, and “often substantially reproduces the original content in its generated output.”
It names ChatGPT directly, calling the problem “particularly the case with a consumer-facing chatbot, like ChatGPT, that relies on RAG techniques to respond to queries.”
Scraping Is Treated Separately
The brief argues that scraping itself should not count as a copyright use at all. “Scraping is simply a means of acquiring copies of works, but acquisition in and of itself is not a use of those works,” it says.
It does not call scraping harmless. A footnote says scrapers should avoid overburdening websites, and offers a comparison: “a public park can remain free and open while preventing someone from driving a container truck through it.”
Wikimedia’s statement adds that scraping of its projects “must be done sustainably and in line with our public API rate limits or through our Enterprise API offerings.”
The Licence Behind the Argument
Most Wikipedia text is licensed under Creative Commons CC BY-SA 4.0, which requires credit to the original creators and that adaptations stay under the same licence. The brief says RAG copies are a problem for Wikimedia when they are used without the attribution and ShareAlike terms of that licence.
Wikimedia’s statement sets out what it wants: “ensuring that uncredited use of the content we host is limited to model training”, with any use for RAG and similar grounding techniques resulting in proper attribution.
Chief executive Bernadette Meehan said: “Expanding access to knowledge through AI is a powerful goal, but it must respect the people, contributors, and communities that make reliable knowledge possible.”
Why the Split Matters for AI Search
As we read it, the split moves the licensing question from the training run to the answer itself. If a court adopted it, the cost of using a publisher’s text would attach to each retrieval at query time, not to a one-off training corpus.
Wikimedia’s statement says Wikipedia holds more than 67 million articles across 300 languages, and calls it one of the highest-quality datasets used to train LLMs.
The Structural Read
As we read the filing, the brief backs neither side’s all-or-nothing position: it accepts the training use the Times challenges and contests the retrieval use OpenAI defends.
As we read it, the licence is the lever: Wikimedia’s statement says it wants uncredited use limited to model training, with retrieval uses giving attribution under CC BY-SA 4.0.
Wikimedia Foundation proposed amicus brief, 29 October 2026
“Scraping is simply a means of acquiring copies of works, but acquisition in and of itself is not a use of those works.”
Three Implications
FOR AI ANSWER ENGINES The brief singles out a consumer-facing chatbot that relies on RAG, the step that copies and pastes web text into answers at query time.
FOR PUBLISHERS Wikimedia’s statement says companies offering RAG should respect each content platform or publisher’s licensing requirements.
FOR SCRAPERS Wikimedia’s statement says scraping of its projects must follow its public API rate limits or go through its Enterprise API offerings.
The filing is a request for leave, and the documents we read do not show a ruling on it. The brief is Wikimedia’s argument, not a finding by the court. OpenAI’s and The New York Times’ responses to it are not in the documents we read.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Wikimedia, which calls its sites one of the most critical sources of data for AI training, is asking Judge Stein to treat training as likely fair and ChatGPT-style retrieval as likely not fair. The brief backs neither the Times nor OpenAI; it asks for a stage-by-stage test.
A note on sourcing. This piece is based on the Wikimedia Foundation’s filing in the Southern District of New York (ECF 2361 and its exhibit, the proposed brief) and the Foundation’s statement of 30 October. Quotations are verbatim from those documents.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
