Wikimedia Urges Middle Ground in AI Copyright Fight
The Wikimedia Foundation has asked a federal court weighing The New York Times’ copyright case against OpenAI and Microsoft to avoid a sweeping ruling on artificial intelligence and fair use. In an amicus brief, Wikipedia’s nonprofit operator says training models on copyrighted material can qualify as fair use, while AI services that reproduce works for users may need permission from rights holders.
Wikimedia argues that treating all AI use of copyrighted material as infringement could restrict new technologies and tools useful to Wikipedia editors. It says language models can help identify vandalism, copyright violations and doxxing, as well as articles needing improvement. Wikipedia content is also a major source of AI training data, the foundation says.
But Wikimedia warns that blanket permission could weaken the sources and communities that make reliable information available. It says AI systems can draw on freely licensed Wikipedia material without following the Creative Commons terms, including requirements to credit sources and keep adaptations under the same license.
The foundation’s proposed distinction is whether a system creates something meaningfully different or instead supplies material that substitutes for the original and diverts readers. It supports scraping and model pre-training when the result is new and different, but says reproducing existing material in response to queries should require a license when it replaces engagement with the source.
The Times alleges that OpenAI infringed its copyright by using material from its online publication to assemble model-training data. OpenAI and Microsoft argue that training and deploying models based on published material is fair use. A decision in the case could influence more than a dozen similar lawsuits; a summary judgment is expected in the near future.
AI Copyright Ruling Must Protect Knowledge and Its Creators
The Wikimedia Foundation is pressing a New York court to reject a blanket fair-use defense that could let AI companies reproduce published work without respecting the people and communities behind it. In its brief in The New York Times’ case against OpenAI and Microsoft, Wikimedia says model training can be fair use, but AI output that substitutes for original works should require a license.
That distinction matters to Wikipedia, whose material is used to train AI systems and is licensed under terms that require credit and impose conditions on adaptations. Wikimedia says systems can draw on its freely shared knowledge without honoring those terms, while also diverting readers from the sources and communities that sustain open information.
The foundation does not call for shutting down AI training. It says models can help Wikipedia editors spot vandalism, copyright violations and doxxing, and identify articles needing attention. It supports training that produces something new and different from its sources.
But OpenAI and Microsoft’s broad fair-use position risks treating the use of others’ work as permission to deliver that work back to users. Wikimedia argues that when AI responses replace engagement with original material, licensing should apply. The Times alleges OpenAI used material from its online publication to compile data for model training; the companies say their uses are fair.
A summary judgment is expected in the near future, and the case could set a precedent in more than a dozen similar lawsuits. The stakes, Wikimedia argues, include whether free knowledge and the contributors who create it can remain viable alongside AI.
Wikimedia Warns Against Copyright Rules That Could Choke AI
The Wikimedia Foundation is warning a New York court not to let copyright claims sweep broadly enough to block transformative AI development. In a brief in The New York Times’ case against OpenAI and Microsoft, the nonprofit behind Wikipedia supports fair use for training models on copyrighted material when they create something new and different from their sources.
Wikimedia says an overly restrictive ruling could obstruct useful tools and open-source technologies. Language models can assist Wikipedia editors in finding vandalism, copyright violations and doxxing, and in locating articles that need improvement. The foundation also says Wikipedia has become a major source of AI training data.
The foundation’s position is not an endorsement of unrestricted output. It says AI companies should need licenses when systems reproduce material in ways that substitute for original works and displace readers. It also argues that services using Wikipedia content should respect its Creative Commons terms, including attribution and conditions for adaptations.
The Times alleges that OpenAI infringed copyright by scraping its site for training datasets. OpenAI and Microsoft say model training and deployment based on published material fall within fair use. Wikimedia cautions that either an overly narrow or overly broad rule could damage access to knowledge: one could constrain useful innovation, while the other could undermine the open sources AI depends on.
A summary judgment is expected in the near future. The case is one of more than a dozen similar suits and could shape how courts apply copyright law to AI.
Reported by The Verge, Engadget.
