维基媒体提交法庭之友简报，主张区分 LLM 训练与输出侵权

The Wikimedia Foundation is staking out a middle ground in a group of landmark AI copyright cases, pushing courts to legally protect large language models (LLMs) training on copyrighted work — but not chatbots regurgitating that work to users. An overly black-and-white ruling, it says, could either deprive Wikipedians of useful AI-based tools or endanger the open web.

In an amicus brief filed Friday, the nonprofit behind Wikipedia weighed in on a set of lawsuits pitting publishers including The New York Times against OpenAI and Microsoft, taking neither side. “Wikimedia is concerned that the parties’ positions in this case push for overbroad and simplified applications of the fair use doctrine” around AI, the organization says in its legal filing. A nuance-free victory for publishers, it fears, could let copyright holders “gatekeep” even innovative and transformative uses of their work. On the other hand, a victory for AI companies could constitute “a green light to snub licensing terms and fragment the flow of knowledge between human authors and audiences.”

“Wikimedia is concerned that the parties’ positions in this case push for overbroad and simplified applications of the fair use doctrine”

The ruling could be existential for Wikipedia. Wikimedia websites “have become one of the most critical sources of data for all artificial intelligence and machine learning training,” the company says in the filing. But these same systems siphon away visitors, undermining the communities that maintain these sites. Wikimedia argues that they’re also violating copyright: services like ChatGPT reproduce text that’s licensed freely under Creative Commons, it says, but don’t adhere to the terms of that license.

At the same time, it says, the large language models that power these systems are useful to Wikipedia: they can help editors fight vandalism, detect copyright violations and doxxing, and look for articles that need improvement, among other uses.

“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 council 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.”

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

The Timesargues for a more expansive definition of fair use as it applies to LLMs. It alleges, for example, that OpenAI infringed on its copyright by scraping its website to build training datasets. OpenAI and Microsoft both contend that using LLMs constitutes fair use. OpenAI has said that regurgitating work “is a rare bug.” Microsoft, in a September filing, says that expert analysis found that Copilot rarely reproduces full sentences from news content used to ground the model, which it said, “hardly undermines the transformative purpose of LLM training.”

Wikimedia proposes a compromise. Scraping websites shouldn’t be considered copyright infringement, it argues, and using copyrighted works to pre-train LLMs without a license should constitute fair use when they provide “something new and different from the original sources,” according to a blog post. But if these models pull copies of material in response to user queries, then deliver it to users “while supplanting readers’ engagement with those original works,” that should require a license. The key difference is that in the latter case, AI output is simply substituting for the original work, and doesn’t transform it enough to qualify as fair use.

In those cases, Wikimedia says that AI companies should respect licensing requirements to use such information. For Wikipedia, this would mean crediting the original source of information under its Creative Commons license, which also requires “any remix or adaptations” to remain under that same license.

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