Wikimedia Brief Says AI Training Can Be Fair Use, but Reproduced Output Should Need a License

The Wikimedia Foundation, taking neither side, filed an amicus brief in publishers' copyright suits against OpenAI and Microsoft, including The New York Times. It argues that scraping websites should not count as infringement, and that pre-training large language models (LLMs) without a license should be fair use when models offer something new. Reproducing material in user-facing answers should require a license.