# Wikimedia asks court to assess AI training and RAG separately

Published 2026-10-30 · AI.info Team · AI.info

> The Wikimedia Foundation’s proposed court brief argues that copyright analysis should distinguish general-purpose AI model training from the use of retrieved material in chatbot responses. It says training may qualify as fair use, while some retrieval-augmented generation uses should respect publishers’ licensing terms.

The New York Times alleges that [OpenAI](https://ai.info/companies/openai) and [Microsoft](https://ai.info/companies/microsoft) infringed its copyrights; the companies argue that using publicly available material to develop AI systems qualifies as fair use. The [Wikimedia Foundation](https://upload.wikimedia.org/wikipedia/commons/2/21/26.10.29_WMF_motion_and_amicus_In_re_OAI.pdf) is asking a federal court in New York to assess different uses of copyrighted works separately, rather than apply one rule to every stage of an AI system. Its proposed amicus brief distinguishes general-purpose model training from retrieval-augmented generation, or RAG.

## Wikimedia’s position in 12 lawsuits

On October 30, 2026, the Wikimedia Foundation announced that it had filed an amicus brief in *In re: OpenAI, Inc. Copyright Infringement Litigation*, a Southern District of New York case combining 12 lawsuits against OpenAI. The filing includes a letter motion dated October 29 requesting permission to submit the brief in support of neither party. The litigation includes The New York Times’ 2023 suit against OpenAI and Microsoft.

Wikimedia is not taking either side. Instead, it asks the court to consider how copyrighted works are used at distinct stages of AI development and operation. The motion states that the proposed brief contains 5,831 words, within the court’s 6,000-word limit.

> “Each distinctive use should be assessed with independent fair-use analyses.”
>
> The brief’s authors—Stephen LaPorte, Wikimedia Foundation general counsel; Jacob Rogers, associate general counsel; and Aly Marino, legal counsel—write.

## Why training and RAG get different analyses

For pre-training a general-purpose language model, Wikimedia argues that copying works is likely fair use. The brief says training is transformative when it produces a model that is new and different from the source works. It argues that copies used in pre-training do not themselves substitute for original works: they are used to analyze syntax and structure and adjust the model’s understanding of relationships between words, rather than reproduce the works’ expressive content in a substantially similar way.

The brief cites the Supreme Court’s 2023 decision in *Andy Warhol Foundation v. Goldsmith* in support of applying fair-use analysis separately to distinct uses. It urges the court to distinguish using copied works to train a general-purpose model from later uses of copies in an AI product.

RAG, the brief argues, presents a different question. A RAG system can search the web or another database, identify and copy relevant material, and place it in a model’s context when answering a user. Wikimedia argues that using retrieved works in ways that reproduce or supplant the originals is likely not fair use. It asks the court to require licensing terms to be respected for such uses, including the Creative Commons terms that apply to Wikimedia content.

## Licenses, attribution and scraping

Wikipedia contributors publish their work under Creative Commons CC BY-SA terms, which require credit to original creators and require adaptations to remain under the same or a compatible license. Those terms matter when AI products retrieve Wikimedia content to answer questions, Wikimedia says. Its proposed distinction is between model pre-training, which it argues may be fair without meeting licensing terms, and RAG uses that reproduce source material.

The filing also distinguishes scraping—the automated process of making copies from websites—from later uses of those copies. Wikimedia argues that scraping itself should not determine the fair-use analysis; courts should examine what is done with copies after they are acquired. The brief notes that scraping can raise other issues, including possible violations of site terms or statutes, apart from copyright fair use.

An overly restrictive fair-use rule could limit access to material used to develop open-source AI technologies, according to the Foundation. The Foundation also warns that a blanket fair-use rule could encourage publishers to put more information behind barriers, reducing sources available to Wikipedia volunteers. Wikipedia has more than 67 million articles in 300 languages, and the Foundation says nearly 250,000 volunteers contribute each month.

Wikimedia’s proposed filing asks the court to assess each distinct use of copied works, including pre-training and RAG. It argues that this approach could leave room for AI development while protecting works when chatbot responses reproduce or replace the originals.

## Sources

1. Original source · Court record [Wikimedia Foundation court filing, hosted on Wikimedia Commons](https://upload.wikimedia.org/wikipedia/commons/2/21/26.10.29_WMF_motion_and_amicus_In_re_OAI.pdf) upload.wikimedia.org

Other pages read

2. [Wikimedia Foundation](https://wikimediafoundation.org/news/2026/10/30/wikimedia-foundation-amicus-brief-new-york-times-v-open-ai-case/) wikimediafoundation.org

## Related stories

- [Wikimedia Finds Unauthorized Activity It Links to OpenAI Agents](https://ai.info/news/wikimedia-openai-agent-unauthorized-activity)
- [Third Circuit Affirms Thomson Reuters’ Win Against Ross Intelligence](https://ai.info/news/third-circuit-affirms-thomson-reuters-ross-intelligence)
- [MLPerf Adds RAG and Edge-Agent Tests to v6.1 Results](https://ai.info/news/mlperf-adds-rag-edge-agent-tests-v6-1-results)
- [Four Entertainment Unions Challenge OpenAI’s Fair-Use Defense](https://ai.info/news/entertainment-unions-challenge-openai-fair-use-defense)
- [Apple Researchers Report Faster Retrieval With Small Neural Networks for Maximum Inner Product Search](https://ai.info/news/apple-neural-networks-faster-rag-retrieval)

## Explore

[More articles](https://ai.info/news)

---

Canonical: https://ai.info/news/wikimedia-court-separate-ai-training-rag
