Wikimedia brief splits fair use from chatbot answers that replace sources

Wikimedia told a court unlicensed LLM pre-training can be fair use, but answers that replace the source need a license.

The Wikimedia Foundation filed an amicus brief in publisher lawsuits against OpenAI and Microsoft, including a case brought by The New York Times, and took neither side. It argues scraping and unlicensed pre-training should be fair use when models add something new, while answers that copy works and replace reader visits should require a license. Wikimedia says Wikipedia is a major training source, chatbots siphon visitors, and tools such as ChatGPT reproduce Creative Commons text without required credit and the same license on adaptations. It warns a publisher win could gatekeep transformative uses and an AI-company win could ignore licensing and fragment knowledge. The foundation also says LLMs help editors spot vandalism, copyright violations and doxxing.