Wikimedia Urges Fair-Use Shield for AI Training, Opposes 2nd-Step Article Regurgitation
Updated
Updated · The Verge · Oct 30
Wikimedia Urges Fair-Use Shield for AI Training, Opposes 2nd-Step Article Regurgitation
3 articles · Updated · The Verge · Oct 30
Summary
An amicus brief filed Friday by the Wikimedia Foundation asks courts to treat scraping and pretraining on copyrighted works as fair use, but to require licenses when chatbots reproduce protected text for users.
Wikimedia says a binary ruling in the New York Times-led cases against OpenAI and Microsoft could either let copyright holders block transformative AI uses or let AI firms ignore licensing terms and siphon traffic from knowledge sites.
Wikipedia’s operator argues the distinction is whether AI output transforms source material or substitutes for it; if a model delivers copied passages that replace reading the original, fair-use protection should not apply.
The filing says the stakes are existential for Wikipedia because its sites are major training sources for AI systems even as those same systems divert visitors, though Wikimedia also uses LLMs to fight vandalism, spot copyright issues and improve articles.
OpenAI and Microsoft maintain LLM use is fair use and say verbatim reproduction is rare, while Wikimedia frames its position as a middle path to protect both open-source AI development and the sustainability of free knowledge.
Why is Wikimedia backing AI training in court while warning that regurgitated answers could damage Wikipedia, publishers, and the future of open knowledge?
If AI can learn from copyrighted articles but not repeat them, where should courts draw the line between fair use and digital substitution?