Wikimedia urges the OpenAI copyright court to treat AI training as fair use but not chatbot retrieval

The Wikimedia Foundation, the nonprofit that hosts Wikipedia, has asked the federal court handling the combined copyright cases against OpenAI to judge two uses of copied works separately: training AI models, and the way chatbots copy web pages into their answers. In a motion filed on October 29, 2026, and announced on October 30, the foundation asked Judge Sidney H. Stein of the Southern District of New York for leave to file a brief in support of neither party in In re: OpenAI, Inc. Copyright Infringement Litigation. It says the parties are pushing for “overbroad and simplified applications” of fair use, when each use of the copied works should get its own analysis.

The case combines 12 lawsuits against OpenAI, the best known of them The New York Times Co. v. Microsoft and OpenAI, which the Times filed in 2023, according to the foundation. In that suit, the foundation says, the Times alleges that any use of its content by OpenAI and Microsoft, including to train AI models, violates copyright law. The court will rule on whether AI companies' use of copyrighted content to develop large language models is fair use.

Training on one side, retrieval on the other

On training, the brief reaches the same answer as OpenAI and Microsoft, which argue that training and deploying models on publicly available data is fair use. Drawing on its own experience building and using machine-learning tools, Wikimedia argues that training a general-purpose language model likely puts the copied works to a new, transformative purpose and does not in itself hurt their market by standing in for them, which makes it likely fair use.

On retrieval-augmented generation, or RAG, it reaches the opposite answer. In the brief's account, a chatbot such as ChatGPT fetches text from sites like Wikipedia or The New York Times, feeds the passages that best match a user's question into the model along with the question, and often repeats much of the original in its reply. Where a service uses copies for the same purpose as the originals and takes the place of reading them, the brief says, that is likely not fair use. It asks the court to rule that “OpenAI's use of copies in ChatGPT's RAG feature is not fair and must abide by the terms of the relevant licenses.”

Scrapers, the brief adds, should still follow the rules of the road and not overburden websites: a public park “can remain free and open while preventing someone from driving a container truck through it.”

Why the license matters to Wikipedia

Most written contributions to Wikimedia's projects are published under the Creative Commons CC BY-SA 4.0 license, which requires credit to the original creators and that adaptations be shared under the same terms. Wikimedia argues that when chatbots use its content without following those terms, people are steered away from the projects and have less reason to contribute. The foundation's post says model training should be the only uncredited use of the content it hosts, and that scraping should stay within its public API rate limits or go through its Enterprise API offerings.

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

The load is not hypothetical, by the foundation's account. In April 2025 it said bandwidth for downloading multimedia had grown 50% since January 2024, largely from bots scraping Wikimedia Commons, and that at least 65% of its most resource-consuming traffic came from bots, although bots made up about 35% of pageviews. On October 5, 2026, it said it had found activity by what it called “rogue” OpenAI agents on its sites, including unapproved edits, mostly in sandbox areas, failed attempts to misuse a note-taking tool, and millions of automated requests that may have contributed to a partial outage of the Wikidata Query Service in May. It said it found no evidence that its systems or data had been compromised.

What happens next

The brief is a proposed one. The court has broad discretion over whether to accept it. Its order of August 7, 2026, later modified on September 24, directs would-be amici to seek leave by motion and sets a 6,000-word limit; Wikimedia's brief runs to 5,831 words. Wikipedia, the foundation says, has more than 67 million articles across 300 languages, with nearly 250,000 volunteers contributing each month.