Wikipedia's parent tells court: either extreme in NYT v. OpenAI threatens the open web

The Wikimedia Foundation has filed an amicus brief in The New York Times' copyright lawsuit against OpenAI and Microsoft, arguing that an extreme ruling in either direction could threaten the open web. The organization behind Wikipedia isn't picking a side outright. It's proposing a middle path, and the timing matters: a summary judgment in the case is expected in the near future, and the decision could set an important precedent for more than a dozen similar cases now working through the courts.

By Sophie Lindqvist

Why is Wikipedia's parent in this fight at all?

The New York Times sued OpenAI and Microsoft over whether their use of Times-published content — for any purpose, including AI model training — violates copyright law. The Times says companies like OpenAI shouldn't get free rein over its published content. OpenAI and Microsoft counter that training and deploying AI models on published data is necessary for technological development, and therefore falls within the definition of fair use.

The Wikimedia Foundation sees danger on both ends of that argument. Its brief agrees with Microsoft and OpenAI on one point: a much more restrictive version of fair use could inhibit tech innovation. But it also warns that a ruling in favor of blanket fair use could lead to a closed web, one where knowledge is weakened due to a "lack of available secondary source materials."

That second concern isn't abstract for Wikipedia. As the brief itself notes, Wikipedia is considered one of the highest-quality datasets used to train large language models. The encyclopedia ecosystem feeds the very technology being litigated, and it also depends on a web full of sources it can cite.

Takeaway: Wikipedia's parent has skin in the game on both sides — as a training dataset and as a project built on open sources.

What does the brief actually propose?

The Foundation's alternative position applies fair use to two specific use cases for AI companies.

• Certain model training scenarios. The Foundation argues that scraping copyrighted content should be considered fair use when it's used in "training LLMs that are something new and different from the original sources." The test, in other words, isn't whether the data was copied — it's what the resulting model does with it and whether it transforms the material into something genuinely distinct.
• Retrieval-augmented generation, or RAG. This is a technique that optimizes the output of large language models by referencing authoritative knowledge bases outside the model's training data. Rather than relying only on what it memorized during training, an LLM using RAG pulls in external sources at answer time to ground its responses.

The support for RAG comes with a clear limit. The Foundation does not support RAG when an LLM is regurgitating pre-existing articles without attribution. Pulling from an authoritative knowledge base to improve an answer: acceptable, in the Foundation's framing. Reproducing someone else's article with no credit: not acceptable.

Takeaway: the brief draws its line at transformation and attribution, not at copying itself.

What did the Foundation's general counsel say?

Stephen LaPorte, the Wikimedia Foundation's General Counsel, put the stakes directly in the brief:

"A ruling that does not apply fair use principles carefully could threaten the sustainability of free knowledge, and stifle the development of open source technologies, ultimately impacting access to information for everyone in both cases."

LaPorte also addressed why the court should care about Wikipedia specifically:

"The court needs to understand that Wikipedia is created through a collaborative process that requires a strong foundation of open and accessible human knowledge to thrive."

That's the mechanism worth pausing on. Wikipedia's articles aren't produced in a vacuum. They're assembled collaboratively, and each one rests on citations to secondary sources — news coverage, books, research — that volunteers can consult. If AI companies vacuum up that source material without limits, and if publishers respond by locking content away, the foundation under the encyclopedia gets thinner. The Foundation's argument is that a careless ruling could produce that outcome from either direction.

Takeaway: the Foundation is defending the supply chain of free knowledge, not one company's business model.

Why does this case matter beyond the Times and OpenAI?

The Times v. OpenAI case is just one of over a dozen similar cases making their way through the courts. What distinguishes it is its potential to set an important precedent — the reason the Wikimedia Foundation, which isn't a party to the lawsuit, bothered to file a friend-of-the-court brief at all.

The core legal question is narrow on its face: does using published content to train and deploy AI models fall within fair use? The answer will shape how AI companies access data, how publishers protect and license their work, and whether open-source projects — which can't easily negotiate licensing deals the way large companies can — still have a path to competitive models.

The Foundation's brief hints at that last point. Its warning that a bad ruling could "stifle the development of open source technologies" reflects a practical worry: if training on copyrighted content requires payment or permission across the board, only the biggest players will be able to afford building models at all.

Takeaway: the ruling will help decide who gets to build AI, not just who owns the data.

What happens next?

A summary judgment in the lawsuit is expected in the near future. Whatever the court decides, the Wikimedia Foundation's filing ensures that at least one voice in the proceedings is arguing against both maximalist positions — no blanket permission for AI companies, and no crushing restriction on training either.

The brief's two-pronged test, fair use for transformative training and for attribution-respecting RAG, offers the court a concrete middle route. Whether the judge takes it will shape not just this case, but the dozen-plus others queued behind it, and with them, the terms on which the open web operates.