Wikipedia Urges Split on AI Training Copyright
The nonprofit's amicus brief asks a federal court to treat model training and retrieval-based generation as legally distinct in a pending case.
Updated on Nov. 2, 2026 in Artificial Intelligence
Live Poll
Should AI companies have to license copyrighted content before using it for training?
The Wikimedia Foundation filed an amicus brief on Friday in federal court in the Southern District of New York, in the copyright case The New York Times brought against OpenAI and Microsoft in 2023. The development is a court filing in a pending case, not a shipped product or settled ruling.
Why it matters
The New York Times lawsuit accuses OpenAI and Microsoft of using its copyrighted articles to develop and power AI systems, while the defendants argue that training on publicly available material is fair use. Wikimedia's brief, filed as a non-party, proposes separating model training from retrieval-augmented generation, a pattern in which systems fetch and reproduce existing web content, because the two may warrant different legal treatment.
The filing relies on the scale of Wikipedia's content base, with more than 67 million articles in roughly 300 languages and about 250,000 monthly volunteer contributors. Whether the court accepts the brief's training-versus-retrieval distinction remains unresolved, and the brief is an argument rather than a ruling.
The players
Wikimedia Foundation
The nonprofit that operates Wikipedia, whose open-licensed encyclopedia of more than 67 million articles is widely used as training data for language models.
The New York Times
The publisher that sued OpenAI and Microsoft in 2023, alleging use of its copyrighted articles to develop and power AI systems.
OpenAI
The AI developer behind the GPT family of large language models, named as a defendant in the New York Times copyright suit.
Microsoft
The technology company and major investor in OpenAI, named as a co-defendant in the New York Times copyright suit.
The details
Large language models are AI systems trained on enormous collections of text to predict and generate language. The Wikimedia brief distinguishes that training step from retrieval-augmented generation, a design in which a system looks up existing web pages at the time of a query and reproduces material from them in its response. The foundation argues that resulting models can be meaningfully distinct from their source material, and that retrieval-based systems should not be presumed to fall under fair use. Wikipedia's content is licensed under Creative Commons CC BY-SA 4.0, a license that permits reuse with attribution and share-alike terms.
Timeline
2023: The New York Times sued OpenAI and Microsoft.
Friday: The Wikimedia Foundation filed its amicus brief in federal court.
The Tech Race
The Wikimedia filing enters a contest over whether AI developers can lawfully use publicly available text, with the New York Times case as the lead test. The outcome will shape how companies source training data for the systems competing across the field.
The brief does not change how any AI product works today, since the case is still pending and no ruling has been issued. For people who write, publish, or edit online, the eventual answer will determine whether publicly posted text can be used to train commercial models without licensing.
The takeaway
Watch for whether the Southern District of New York addresses the training-versus-retrieval distinction in any ruling on fair use. That decision would be the first signal of whether retrieval-augmented systems face a different legal standard than model training.
Further reading
For the wider legal fight over training data, see the Artificial Intelligence coverage on We are the Frontier.
Source note: This article includes information reported by Benzinga.
Live Poll
Should AI companies have to license copyrighted content before using it for training?
