Wikimedia warns AI copyright ruling could break the open web

The Wikimedia Foundation filed an amicus brief on Friday arguing that training large language models on copyrighted works should count as fair use — but that chatbots regurgitating those works to users should not.

The filing lands in the middle of a set of landmark lawsuits that pit publishers including The New York Times against OpenAI and Microsoft. Wikimedia takes neither side. Instead, it warns that whichever way the court rules, an overly black-and-white decision could do real damage: either to Wikipedia's editor communities or to the open web itself.

"Wikimedia is concerned that the parties' positions in this case push for an overbroad and simplified applications of the fair use doctrine," the organization says in its legal filing. The quote captures the core of its argument: fair use is a balancing test, and both camps want a clean win.

Why does Wikipedia care about an OpenAI lawsuit?

The stakes are existential for the nonprofit behind Wikipedia. Wikimedia websites "have become one of the most critical sources of data for all artificial intelligence and machine learning training," the filing states. In plain terms: AI systems are built, in large part, on Wikipedia's text.

But the relationship cuts both ways. Those same systems siphon away visitors, which undermines the volunteer communities that keep the sites alive. Wikimedia also argues that current AI services violate copyright in a specific, concrete sense. Services like ChatGPT reproduce text that's licensed freely under Creative Commons, it says, but they don't adhere to the terms of that license.

At the same time, Wikimedia can't pretend the technology is purely a threat. Large language models are useful to Wikipedia's editors. The filing notes they help fight vandalism, detect copyright violations and doxxing, and find articles that need improvement. So Wikimedia isn't asking courts to shut the technology down — it's asking for a rule that lets the useful parts survive.

The takeaway: Wikipedia is both a supplier of training data and a potential victim of the products built on it, and it wants the law to reflect that split.

What are the parties actually arguing?

The New York Times argues for a more expansive definition of infringement. It alleges OpenAI infringed its copyright by scraping its website to build training datasets — the raw data-gathering step itself, in the Times' view, crosses the line.

OpenAI and Microsoft both contend that using works to train LLMs constitutes fair use. On the regurgitation problem, OpenAI has said that reproducing work "is a rare bug." Microsoft went further in a September filing, citing expert analysis that found Copilot rarely reproduces full sentences from news content used to ground the model. That, Microsoft argued, "hardly undermines the transformative purpose of LLM training."

So the publishers want scraping treated as infringement, and the AI companies want nearly everything folded into fair use. Wikimedia's position is that both framings miss the point.

What compromise does Wikimedia propose?

Wikimedia draws the fair use line between two very different behaviors.

• Scaping and pre-training: Scraping websites shouldn't be considered copyright infringement, Wikimedia argues. Using copyrighted works to pre-train LLMs without a license should constitute fair use when the models provide "something new and different from the original sources," according to a blog post accompanying the brief.
• Serving copies to users: If a model pulls copies of material in response to user queries and delivers them "while supplanting readers' engagement with those original works," that should require a license.

The mechanism here is substitution. If AI output simply replaces the original work — a reader asks a chatbot and never visits the source — the use isn't transformative enough to qualify as fair use. If the model digests the work and produces something genuinely new, it is.

In the licensing cases, Wikimedia says AI companies should respect the terms attached to the information. For Wikipedia's content, that means crediting the original source under its Creative Commons license, which also requires that "any remix or adaptations" stay under that same license. Those aren't onerous terms; they're the terms Wikipedia itself operates under.

The takeaway: Wikimedia wants a rule that judges what the AI output does to the original work, not what the training pipeline touched.

What does Wikimedia's leadership say?

The filing comes with statements from two of the Foundation's top figures, and both are worth reading in full.

"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," Wikimedia Foundation general counsel Stephen LaPorte says in a statement. "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."

CEO Bernadette Meehan frames the position as defense rather than neutrality. "We are taking a position in this case to defend the free and open human knowledge that Wikipedia depends on," Meehan says in a statement. "Expanding access to knowledge through AI is a powerful goal, but it must respect the people, contributors, and communities that make reliable knowledge possible. Finding this balance is both a mutual benefit and a shared responsibility for the entire information ecosystem."

Why does this matter beyond Wikipedia?

These cases — the Times v. OpenAI chief among them — will set the boundaries of fair use for an entire industry. A nuance-free victory for publishers, Wikimedia fears, could let copyright holders "gatekeep" even innovative and transformative uses of their work. A clean victory for AI companies, it warns, could constitute "a green light to snub licensing terms and fragment the flow of knowledge between human authors and audiences."

Both failure modes hit the same target: the freely accessible web. Publishers gating their archives and AI firms ignoring licenses both narrow the pool of openly available knowledge. Wikipedia, built entirely on freely licensed contributions, sits at the intersection of those pressures.

Amicus briefs don't decide cases, but they can shape how judges frame the questions. Wikimedia's contribution here is a workable middle path: protect training, regulate substitution, and enforce license terms — including Creative Commons — when AI products serve copyrighted material to users. How much of that framework survives contact with the court's ruling will matter for anyone who publishes, trains, or builds on the open web.