Wikimedia Foundation Advocates Nuanced Approach in Landmark AI Copyright Cases

The Wikimedia Foundation, the nonprofit organization behind Wikipedia, has entered a critical debate surrounding landmark AI copyright cases by staking out a middle ground. In an amicus brief filed Friday, the Foundation urged courts to legally protect large language models (LLMs) when they are training on copyrighted work, but to differentiate this from chatbots that regurgitate that work directly to users. This position comes amidst lawsuits pitting publishers, including The New York Times, against AI developers like OpenAI and Microsoft, with Wikimedia expressing concern that the parties' positions advocate for "overbroad and simplified applications of the fair use doctrine" around AI.

Wikimedia fears that an overly black-and-white ruling could have severe consequences. A victory for publishers might allow copyright holders to "gatekeep" even innovative and transformative uses of their work, while a win for AI companies could "constitute a green light to snub licensing terms and fragment the flow of knowledge between human authors and audiences." The organization stated that the ruling could be "existential for Wikipedia," noting that Wikimedia websites are "one of the most critical sources of data for all artificial intelligence and machine learning training." However, these same AI systems are also observed to siphon away visitors, thereby undermining the communities that maintain these sites.

The Foundation argues that services like ChatGPT often reproduce text licensed freely under Creative Commons but fail to adhere to the terms of that license, constituting a copyright violation. Conversely, Wikimedia acknowledges the utility of large language models for Wikipedia itself, as they can assist editors in combating vandalism, detecting copyright violations and doxxing, and identifying articles needing improvement. Stephen LaPorte, Wikimedia Foundation general council, emphasized, "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."

To navigate this complex landscape, Wikimedia proposes a compromise. It suggests that scraping websites to pre-train LLMs without a license should be considered fair use if the models provide "something new and different from the original sources." However, if these models retrieve and deliver copies of material in response to user queries, "while supplanting readers’ engagement with those original works," then a license should be required. This distinction hinges on whether the AI output transforms the original work or merely substitutes it. The New York Times, for instance, alleges OpenAI infringed its copyright by scraping its website for training, while OpenAI and Microsoft contend their LLM use is fair use, with OpenAI calling regurgitation "a rare bug" and Microsoft citing expert analysis that Copilot rarely reproduces full sentences. Wikimedia CEO Bernadette Meehan affirmed, "We are taking a position in this case to defend the free and open human knowledge that Wikipedia depends on."

What to watch: The court's interpretation of fair use and its implications for the future of AI development and the open web.

Editor's note: A solid report that balances the Wikimedia Foundation's legal stance with the broader context of the AI copyright debate.