Wikipedia's owner asks courts for a middle path on AI and copyright

The Wikimedia Foundation, the nonprofit that runs Wikipedia, has asked courts to take a middle path in the big AI copyright fight. In a brief filed Friday, it says training AI on copyrighted work should generally be allowed. But it says chatbots that hand that work back to users should need a license.

What Wikimedia is asking for

The brief is a friend-of-the-court filing, where an outside group shares its views without joining either side. It was filed in the lawsuits brought by publishers including The New York Times against OpenAI and Microsoft. The core issue is fair use, a legal idea that allows some use of copyrighted material without permission.

• Scraping websites should not count as copyright infringement.
• Training AI without a license should be fair use if the result is something new and different.
• Pulling copies of material for users, in a way that replaces reading the original, should need a license.

Why Wikipedia has a stake

Wikimedia says its sites are among the most important sources of data for training AI. It also says those same AI systems pull visitors away and weaken the volunteer communities that maintain the sites. Wikipedia text is free to reuse under a Creative Commons license. Wikimedia says services like ChatGPT reproduce that text but do not follow the license terms.

That license requires credit to the source. It also requires any remix or adaptation to stay under the same license. Wikimedia says AI companies should respect those rules when they reuse its material.

The group is not against AI. It says large language models, the technology behind chatbots, can help its editors fight vandalism, spot copyright violations and doxxing, and find articles that need work.

What the other sides say

The Times argues for a broader view of fair use and says OpenAI broke copyright law by scraping its website to build training data. OpenAI and Microsoft say using these models is fair use. OpenAI has called copying work back to users a rare bug. Microsoft says expert analysis found its Copilot tool rarely reproduces full sentences from news content.

Wikimedia warns that either side winning outright could cause harm. A total win for publishers could let rights holders block even creative, useful uses. A total win for AI companies could ignore licensing terms and break the flow of knowledge between people who write and people who read.

What happens next

The material does not say when the court will rule or whether it will follow Wikimedia's approach. For now, the filing shows that even a site built on free sharing wants rules attached. The outcome could affect how much AI tools can copy from the free sites you rely on.