Wikimedia asks to file brief in OpenAI copyright case

The Wikimedia Foundation asked a federal judge on October 29 for permission to file a proposed amicus brief in the consolidated OpenAI copyright litigation. The filing urges the court to evaluate each use of copyrighted material separately, a framework that could treat transformative model training as fair use while requiring retrieval systems to comply with applicable licenses when they reproduce source material.

Wikimedia supports neither side in the case. Its proposed brief argues that pre-training a general-purpose large language model is likely fair use when the resulting system is transformative and does not substitute for the works used to train it. By contrast, it says ChatGPT’s retrieval-augmented generation, or RAG, should not qualify as fair use when it copies expressive material into answers that supplant the original source.

RAG lets a system fetch outside material in response to a query and use that material to construct an answer. Wikimedia’s position is narrower than a categorical objection to the technology: its argument focuses on retrieval that reproduces protected material in answers and substitutes for the original source.

The foundation also argues that scraping is a way to acquire copies, not the copyright use that should determine the fair-use question. In its view, courts should examine what developers do with those copies during development and deployment instead of treating acquisition, training and retrieval as one undifferentiated act.

The distinction is especially relevant to Wikipedia because most written contributions to Wikimedia projects use the Creative Commons CC BY-SA 4.0 license. That license requires attribution and requires adaptations to remain under the same terms. In its October 30 public explanation, the foundation said RAG and similar grounding systems using Wikimedia content should comply with those conditions.

Wikimedia described the multidistrict proceeding as a consolidation of 12 lawsuits against OpenAI, including The New York Times’ case. It said an overly restrictive approach to fair use could constrain open-source AI development, while blanket fair use could weaken licensing, attribution and the supply of openly available source material.

The court-stamped document is a motion for leave with the proposed brief attached. The opened record does not establish that the Southern District of New York has granted Wikimedia permission to file it.