Wikimedia Backs Fair Use for AI Training but Not RAG

In an announcement, the Wikimedia Foundation said it filed an amicus brief in consolidated copyright litigation against OpenAI and Microsoft in the Southern District of New York. The filing asks the court to distinguish between using copyrighted works for model training and using them for retrieval augmented generation, or RAG.

The foundation argues that model training can qualify as fair use when a model is new and different from the original sources. Under that standard, developers would not need to meet licensing terms for training material.

For RAG and similar grounding methods, the foundation argues that copying content and returning it through a chatbot should not qualify as fair use. It says providers should follow publisher licenses, including Wikipedia's Creative Commons CC BY-SA 4.0 license, which requires attribution and applies the same license to adaptations. The foundation also says companies scraping Wikimedia projects must follow public API rate limits or use its Enterprise API services.