Wikipedia's parent asks a court not to pick an extreme side on AI and copyright
The Wikimedia Foundation wants judges to treat AI training and AI chatbot answers differently in the case that could reshape copyright law for every AI company.
What happened: The Wikimedia Foundation, which runs Wikipedia, filed a friend-of-the-court brief in a New York federal case combining 12 lawsuits, most prominently The New York Times against OpenAI and Microsoft. Rather than back either side, Wikimedia is urging the judge to avoid a simple, all-or-nothing ruling on whether training AI on copyrighted material counts as fair use. It argues the real answer depends on what happens at each stage of use, not a blanket yes or no.
Why it matters: Wikipedia's more than 67 million articles, written by roughly 250,000 volunteers, are among the most-used training data for AI systems and also power search engines and voice assistants. Wikimedia says a ruling that's too strict could block open-source AI developers from using public content at all, leaving only large corporations able to afford licenses. A ruling that's too loose could let AI companies ignore licensing terms entirely, which Wikimedia says already happens: it accuses chatbots like ChatGPT of reproducing Creative Commons text without the required credit, while also pulling readers away from the original sites.
How it works, plainly: Wikimedia draws a line between two things AI systems do. Training a model on large amounts of text to learn general patterns, it argues, should usually count as fair use because the resulting model is different from any single source. But when a chatbot uses a technique called retrieval-augmented generation to fetch specific content and hand it to a user as an answer, Wikimedia says that's different: it can substitute for reading the original source entirely. In that case, it argues, the AI company should have to honor the source's licensing terms, such as giving credit.
What's next: The case remains before a federal judge in the Southern District of New York, who must weigh competing views: the Times wants a stricter standard that could require licensing for both training and output, while OpenAI and Microsoft argue nearly all their uses of copyrighted material qualify as fair use. OpenAI has called instances of reproducing text a rare bug, and Microsoft says independent analysis found Copilot rarely repeats full sentences from news sources. No ruling has been issued, and Wikimedia's brief is one voice among many the court will consider.