# Wikimedia backs AI training, but draws a hard line at copied text

Published 2026-10-30T09:23:56+00:00 · Updated 2026-10-30T09:23:57+00:00 · By superintelligencenews@gmail.com
Canonical: https://superintelligencenews.com/ai-fields/ai-copyright-wikimedia-fair-use/

> Wikimedia filed a court brief backing fair use for some AI training, but arguing that chatbot outputs copying source text should require permission. The foundation says the law must protect both innovation and open knowledge.

## Key facts

- Wikimedia filed its amicus brief on Friday in the New York Times copyright cases.
- The New York Times has sued OpenAI and Microsoft over alleged copyright infringement tied to AI training and outputs.
- Wikimedia says its sites are among the most critical sources for AI and machine-learning training.
- Stephen LaPorte said, “A ruling that does not apply fair use principles carefully could threaten the sustainability of free knowledge”
- Bernadette Meehan said, “We are taking a position in this case to defend the free and open human knowledge that Wikipedia depends on.”

The Wikimedia Foundation filed an amicus brief on Friday in the New York Times copyright cases against OpenAI and Microsoft, backing a narrow fair-use rule for AI training while opposing chatbot outputs that reproduce source text without permission. The filing matters because it places Wikipedia’s parent organization in a middle position: pro-training, but firmly against AI systems that substitute for original works.

 

## What Wikimedia is asking the court to protect

 

Wikimedia’s brief supports the idea that some model training on copyrighted works can qualify as fair use, especially when the result is a new system that does not copy or replace the original work. At the same time, the foundation says there should be a license requirement when a model serves copied material in response to a user prompt and effectively stands in for the source.

 

That distinction is central to the filing. Wikimedia is not asking the court to shut down AI training across the board. Instead, it is trying to separate the act of learning from the act of regurgitating.

 

### How does Wikimedia describe the line it wants drawn?

 

Wikimedia says scraping should not automatically count as infringement, and it argues that pre-training can be fair use when the model generates something new and different. But the foundation says the legal analysis changes when a system outputs copied material to users in a way that substitutes for the original work.

 

The nonprofit also says those uses should respect Creative Commons attribution and share-alike requirements, which are central to much of the Wikipedia ecosystem.

 

## Why it matters

 

This is more than a dispute between a newspaper and AI companies. Wikimedia is one of the internet’s largest knowledge platforms, and its position signals that major open-knowledge institutions do not all see the copyright fight as a simple choice between unrestricted training and total restriction.

 

The filing highlights a possible middle ground that could shape how courts think about large language models. If judges adopt a rule that is too broad in favor of publishers, Wikimedia warns it could chill transformative uses of copyrighted material. If the court goes too far in favor of AI companies, it could reward ignoring licensing terms and weaken the flow of knowledge that open platforms rely on.

 

For readers, that means the case could affect more than who pays whom. It could influence whether AI tools can keep learning from public web material, how often they must credit or license source text, and whether knowledge sites like Wikipedia continue to benefit from the traffic and volunteer energy that support them.

 

## Why is Wikipedia involved in a copyright case about AI?

 

Wikipedia is involved because Wikimedia says its sites are among the major data sources used for AI and machine-learning training. The foundation has a direct stake in both sides of the dispute: it wants AI developers to be able to learn from open knowledge, but it also wants to stop systems from reproducing its material in ways that violate licensing terms or undercut its own work.

 

The foundation also says AI systems can reduce traffic to its sites, which hurts volunteer communities that build and maintain the content. That concern gives the filing an economic and editorial dimension beyond the pure copyright arguments.

 

## What the filing says about AI and Wikipedia

 

Wikimedia’s brief describes its platforms as one of the most important sources of data for AI and machine-learning systems. At the same time, it says those same systems can reduce visits to the sites that produced the material in the first place.

 

That tension is one reason the foundation is arguing for a narrower rule. It wants the law to preserve the ability of AI systems to learn from public knowledge without letting them become a substitute for the original source.

 

### What uses of large language models does Wikimedia support?

 

Wikimedia says large language models can help editors identify vandalism, copyright problems, doxxing, and articles that need improvement. In other words, it sees AI as a tool for editing and moderation, not just a consumer-facing product.

 

That is an important distinction. The foundation is not arguing that AI belongs outside Wikipedia’s workflow. It is arguing that the technology should assist human judgment rather than replace the work of human contributors and the source material they rely on.

 

## The legal fight behind the filing

 

The amicus brief lands inside The New York Times’ copyright lawsuits against [OpenAI](https://openai.com/) and Microsoft. The Times argues that OpenAI scraped its site to build training datasets and infringed copyright in the process. OpenAI and Microsoft say large language model use is fair use.

 

OpenAI has said regurgitating copyrighted text is a rare bug. Microsoft, meanwhile, has said Copilot rarely reproduces full sentences from news content used to ground the model. Those claims go to the heart of the dispute: whether occasional reproduction is an unavoidable edge case or evidence of broader misuse.

 

## Background: how we got here

 

Fair use is a long-standing doctrine that allows limited use of copyrighted material without permission in certain circumstances. Courts usually weigh the purpose of the use, how much was taken, and whether the use harms the market for the original work.

 

AI training has complicated that familiar test. Large language models are built on huge text datasets, and those systems can sometimes generate outputs that resemble the material they learned from. That makes the legal question different from ordinary reading, quotation, or indexing.

 

Wikipedia adds another layer. Much of its content is published under Creative Commons licenses, which allow reuse but often require attribution and, in some cases, share-alike treatment. Wikimedia says ChatGPT-like systems can violate those terms if they reproduce licensed text without the required attribution or share-alike treatment.

 

The New York Times has sued OpenAI and Microsoft over alleged copyright infringement tied to both AI training and outputs. The filings on each side reflect a broader industry struggle over whether model training itself is legal, and what happens when a chatbot produces a near-copy of protected text.

 

## By the numbers

 

| Item | What the fact sheet says | Why it matters |
| --- | --- | --- |
| Friday | Wikimedia filed its amicus brief on Friday. | It adds a major nonprofit voice to the copyright fight. |
| One of the most critical sources | Wikimedia says its sites are among the most critical sources for AI and machine learning training. | That underscores how central open knowledge is to model development. |
| September | Microsoft’s filing said expert analysis found Copilot rarely reproduces full sentences from news content. | That claim is part of the companies’ defense against infringement allegations. |

 

## How Wikimedia is positioning itself

 

The foundation’s argument is deliberately balanced. It warns that a broad win for publishers could block transformative uses of copyrighted material, but it also warns that a broad win for AI companies could erase the limits that make licensing meaningful.

 

That middle position is also a practical one. Wikimedia depends on human contributors, open licensing, and broad access to knowledge. A legal standard that lets AI systems learn from public text while preventing them from replacing the source fits that mission better than an all-or-nothing ruling.

 

> “A ruling that does not apply fair use principles carefully could threaten the sustainability of free knowledge”

 

That was Stephen LaPorte, general counsel for the Wikimedia Foundation, in remarks tied to the filing. He also said Wikipedia needs open, accessible human knowledge to function well, a point that sits at the center of the brief.

 

> “We are taking a position in this case to defend the free and open human knowledge that Wikipedia depends on.”

 

Those were the words of Bernadette Meehan, Wikimedia’s chief executive, who framed the filing as a defense of contributors and the communities that sustain the platform.

 

## What makes this filing different from other AI copyright arguments?

 

Wikimedia is not asking for a categorical ban on using copyrighted text to train AI. It is also not endorsing the view that anything done in the name of model development should automatically be protected. Instead, the filing tries to preserve a distinction that courts may need to keep intact: training can be lawful, but output that copies the source too closely can still require permission.

 

That position may appeal to institutions that rely on open licensing, but it also gives the court a framework for separating productive machine learning from direct substitution.

 

### What if the court sides too strongly with one side?

 

If the court leans too far toward publishers, Wikimedia says transformative uses of copyrighted material could be blocked. If the court leans too far toward AI companies, it could normalize disregard for licensing terms and weaken the knowledge-sharing ecosystem that open platforms depend on.

 

Either outcome would have consequences beyond the parties in the lawsuit. The ruling could influence how future cases treat training datasets, whether near-verbatim outputs trigger licensing duties, and how much weight courts give to market harm versus innovation.

 

## What happens next

 

The next step is for the court to continue weighing arguments in the New York Times cases against OpenAI and Microsoft. The immediate question is how judges will define fair use when the alleged harm is not just copying at the training stage, but the possibility that a model will surface copied text to users later.

 

Watch for three things:

 

- whether the court treats AI training and AI outputs as separate legal questions;
- whether licensing becomes required when a system returns copied or near-copied material;
- whether other publishers, nonprofits, or AI companies cite Wikimedia’s middle-ground position.

 

The filing also raises a practical question for Wikipedia and similar projects: if AI tools continue to absorb large amounts of open web text, can those platforms preserve traffic, attribution, and volunteer engagement while still supporting machine learning? That answer may depend as much on product design and licensing practices as on the court’s final ruling.

 

## The broader stakes for open knowledge and AI

 

Wikimedia’s intervention shows how high the stakes have become for institutions built on shared information. The foundation is not simply defending its own archives. It is arguing that the legal system should protect both the ability of AI to learn and the rights of creators, volunteers, and communities whose work powers that learning.

 

That argument may prove persuasive precisely because it avoids the extremes. It accepts that AI systems need data, and it accepts that source text cannot be treated as free fuel once a model starts serving it back to users. In a fight often framed as publishers versus tech companies, Wikimedia is insisting the real question is how to preserve both innovation and the public knowledge base that makes innovation possible.

## Frequently asked questions

### What is Wikimedia asking for in the AI copyright case?

Wikimedia is asking the court to allow some AI training on copyrighted works as fair use, while drawing a firm line against models that reproduce source text for users without permission. The foundation says training and output need different legal treatment.

### Why did Wikimedia file a brief in the New York Times case?

Wikimedia filed because it says AI models rely heavily on open knowledge sources like Wikipedia, and the outcome could affect both licensing rules and access to information. The nonprofit wants courts to protect training that creates something new without allowing copied output.

### Does Wikimedia support AI training?

Yes. Wikimedia supports the idea that some training on copyrighted material can qualify as fair use. Its objection is to chatbot behavior that serves copied material back to users in a way that substitutes for the original work.

### How could this ruling affect Wikipedia?

The ruling could affect Wikipedia’s traffic, its volunteer community, and how its Creative Commons content is used by AI systems. Wikimedia says AI can also help editors, but it wants attribution and licensing rules respected when material is reproduced.

### What are OpenAI and Microsoft arguing?

OpenAI and Microsoft argue that large language model use is fair use. OpenAI says regurgitating copyrighted text is a rare bug, while Microsoft says Copilot rarely reproduces full sentences from news content used to ground the model.

## Sources

- [The Verge](https://www.theverge.com/policy/1022880/wikipedia-openai-new-york-times-copyright)