Wikimedia Foundation Takes Middle Ground in OpenAI Copyright Case, Warns of Open Web Risks

Wikimedia Foundation Stakes Middle Position in Landmark AI Copyright Case

The Wikimedia Foundation has entered the legal battle between The New York Times and OpenAI and Microsoft with an amicus brief that rejects both sides' positions on AI copyright and fair use1

1

. Filed with the US District Court for the Southern District of New York, the brief argues that copyright infringement litigation surrounding large language models demands nuanced treatment rather than blanket rulings2

2

. The nonprofit organization behind Wikipedia warns that oversimplified decisions could either allow copyright holders to "gatekeep" transformative use or give AI companies license to ignore licensing requirements entirely1

1

.

AI Model Training Qualifies as Fair Use, Organization Argues

Wikimedia's position distinguishes between different AI applications. The foundation contends that using copyrighted works to pre-train large language models without a license should constitute fair use when they provide "something new and different from the original sources"1

1

. According to the brief, AI model training creates transformative use because the resulting model differs fundamentally from the works it learned from2

2

. The organization argues that scraping websites for LLM training purposes shouldn't be considered copyright infringement in itself1

1

. This stance supports innovation while acknowledging that Wikimedia websites "have become one of the most critical sources of data for all artificial intelligence and machine learning training"1

1

.

Retrieval-Augmented Generation Crosses Copyright Line

While supporting training, Wikimedia draws a firm line at retrieval-augmented generation and similar techniques. When chatbots pull copies of material in response to user queries and deliver it "while supplanting readers' engagement with those original works," that should require a license1

1

. The brief explains that chatbots using retrieval-augmented generation copy text from sites like Wikipedia or The New York Times when users ask questions, feeding relevant passages to the model and often reproducing much of the original in answers2

2

. "While the pre-training of ChatGPT's general LLM was fair use, OpenAI's use of copies in ChatGPT's RAG feature is not fair," Wikimedia stated in the brief2

2

. For Wikipedia content, this means respecting Creative Commons CC BY-SA 4.0 licensing, which requires crediting original authors and keeping adaptations under the same license2

2

.

Traffic Decline Threatens Wikipedia's Volunteer Community

The stakes extend beyond legal theory. Wikimedia reports that human visits to Wikipedia have dropped 8% since January 2024, while multimedia scraping bandwidth has increased 50%3

3

. These AI systems siphon away visitors, undermining the volunteer communities that maintain the sites1

1

. Services like ChatGPT reproduce text licensed freely under Creative Commons but don't adhere to those licensing terms, according to Wikimedia1

1

. When chatbots use Wikipedia content without following attribution requirements, people are steered away from contributing2

2

. General counsel Stephen LaPorte emphasized that "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"1

1

.

Balancing Innovation With Attribution Requirements

Despite concerns about unattributed AI outputs, Wikimedia acknowledges that large language models provide valuable tools for Wikipedia editors. These systems help fight vandalism, detect copyright violations and doxxing, and identify articles needing improvement1

1

. A blanket rule against training could hurt public-interest tools3

3

. CEO Bernadette Meehan stated that "expanding access to knowledge through AI is a powerful goal, but it must respect the people, contributors, and communities that make reliable knowledge possible"2

2

. The foundation wants heavy users to follow public rate limits or pay for its Enterprise service, which brought in $8.3 million in 2024-2025, up 148% year over year from customers including Google, Amazon, and Microsoft3

3

. This represents only about 4% of total income, with donations remaining the core support3

3

.