Skip to content
Sign in
Technology · North America

Wikimedia backs AI training fair use but objects to chatbot reproductions

Wikimedia argues that AI training on copyrighted material can qualify as fair use, while chatbot outputs should not reproduce that material.

AI-assisted synthesis · Based on 5 outlets' reporting · Article text read from 2 of 5 outlets · 1 article captured only in partHow it's made

View timeline · 7 entries
5 outletsLatest reporting 3 min read
Coverage: Too few ratedL 0 · C 1 · R 0 · 4 unrated
Updat3 graphic: Wikimedia Foundation
Updat3 graphic

What happened

Key facts

  • The Foundation says an overly broad fair-use ruling could threaten the open web, free knowledge and access to information (per The Verge, avclub.com).
  • The Foundation says its websites are important sources of data for AI and machine-learning training (per The Verge, avclub.com).
  • The Foundation says chatbots may reproduce material licensed under Wikimedia’s Creative Commons terms without following those terms (per avclub.com, The Verge).
  • Microsoft said in a September filing that expert analysis found Copilot rarely reproduces full sentences from news content used to ground the model, and argued that such reproduction hardly undermines the transformative purpose of large language model training (per The Verge).

The brief

The Wikimedia Foundation has urged the court in The New York Times’ case against OpenAI to distinguish between training AI on copyrighted works and chatbots reproducing those works. The Foundation supports fair-use protection for training, but argues that chatbot outputs should not reproduce copyrighted material in ways that fail to respect its licensing terms.

The filing puts the Foundation’s concern over the consequences of the court’s fair-use approach at the center of its position. The Wikimedia Foundation says a broad ruling could put the open web and the sustainability of free knowledge at risk, while also affecting access to information and open-source technology.

The Foundation’s general counsel, Stephen LaPorte, said careful use of fair-use principles matters to those outcomes. The Foundation also describes a practical tension in its relationship with AI tools.

It says Wikipedia websites have become important sources of data for AI and machine-learning training, while AI systems use contributors’ work and can divert traffic from the encyclopedia. The Foundation further argues that chatbots may reproduce material released under Wikimedia’s Creative Commons licensing without complying with its conditions.

That concern does not amount to opposition to every use of AI at Wikipedia. The Foundation says AI tools have helped editors detect plagiarism and vandalism. Its position, as described in the coverage, is that training and particular reproduced outputs should be treated differently under copyright and licensing rules.

OpenAI has called the reproduction of copyrighted work a “rare bug.” Microsoft, in a September filing, cited expert analysis that Copilot rarely reproduces full sentences from news content used to ground the model. Microsoft argued that this kind of limited reproduction does little to weaken the transformative purpose of large language model training.

The Verge’s account emphasizes the Foundation’s warning about the broader consequences of a court ruling, including risks to free knowledge and open-source development. AV Club foregrounds the Foundation’s account of the trade-off for Wikipedia: AI can draw on contributors’ work and divert visits, but AI tools can also assist editors.

These accounts add different emphasis without presenting incompatible descriptions of the Foundation’s position. The dispute is about how copyright’s fair-use doctrine applies to AI training and outputs in the ongoing case brought by The New York Times against OpenAI.

The filing brings Wikimedia’s interests into that debate: its material is used in AI training, and the Foundation says both the licensing conditions on reproduced content and the usefulness of AI tools to editors matter to its position.

Why it matters

  • Wikipedia contributors’ work is used to train AI systems, while the Wikimedia Foundation says those systems can divert traffic from the encyclopedia—an identified cost to the free encyclopedia’s audience and contribution ecosystem.
  • A court ruling on fair use could affect the Wikimedia Foundation’s stated stakes: the open web, free knowledge, access to information and open-source technology.
  • The Foundation identifies competing practical effects: AI tools may reproduce Creative Commons material without meeting its conditions, while also helping editors detect plagiarism and vandalism.
  • OpenAI calls regurgitation a rare bug; Microsoft says Copilot rarely reproduces full sentences from grounding news content. Those positions frame the significance of reproduction differently from the Foundation’s licensing concerns.

What to watch next

  • Whether the court addresses the Wikimedia Foundation’s requested distinction between AI training and chatbot reproduction in The New York Times’ case against OpenAI.
  • Whether the court’s fair-use analysis takes account of the Foundation’s concerns about Creative Commons conditions and reproduced outputs.
  • Whether OpenAI’s characterization of reproduction as a rare bug and Microsoft’s account of Copilot’s limited reproduction feature in the court’s consideration.
  • Whether the Foundation’s concerns about traffic diversion and its account of AI assistance to editors appear in the court’s treatment of the issues.

How it's being covered

Coverage analysisShow details
5 source articles · 5 unique outlets
Outlet political mix: Too few rated· 1 Center · 4 unrated
Where the outlets are based: Unknown 60% · US/Canada 40%
Outlet political mix
Too few rated
Left
Center
Right
0 outlets
1 outlet
0 outlets
The marker shows coverage balance: each unique outlet counts once. 4 unrated outlets are not included.
Geography
Where the outlets covering this story are based.
5 unique outletsDominant coverage: US/Canada
How we measure coverage

How outlets' accounts compare

1 framing difference · 4 uncorroborated claims · 2 information gaps
The reports repeat the same account with different emphasis; its central claims are not independently confirmed in the reporting Updat3 read.Compared across 2 reports whose article text Updat3 read; 3 other outlets were checked by headline only.
Framing / emphasis differences
Sources emphasise different angles, actors or consequences of the same reporting.
  • The Verge emphasizes the Foundation’s warning that a broad ruling could threaten free knowledge, the open web and open-source technology; AV Club emphasizes the traffic Wikipedia may lose to AI tools and their reported use by editors to detect plagiarism and vandalism.
Updat3: These are differences of emphasis, not disputes about what is reported. They do not show that the reported claims are true.
One-sided or uncorroborated claims
Claims that come from one party or one source family and are not independently corroborated.
  • OpenAI characterizes reproduction of copyrighted work as a rare bug, while Microsoft cites expert analysis that Copilot rarely reproduces full sentences from news content used to ground the model; the sources do not establish a shared measure of how often chatbots reproduce Wikimedia material.
  • The Wikimedia Foundation’s general counsel, Stephen LaPorte, said fair-use principles must be applied carefully to protect free knowledge and access to information.
  • OpenAI has described regurgitation of copyrighted work as a “rare bug.”
  • Microsoft argued that rare reproduction of full sentences “hardly undermines the transformative purpose of LLM training.”
Updat3: Not independently corroborated yet — treat as a claim, not an established fact.
Not established by the reporting
Information the available reporting does not contain or confirm. This is a gap, not a disagreement.
  • The sources do not provide the court’s schedule, a decision date, or the specific legal remedy the Wikimedia Foundation seeks.
  • The sources do not quantify traffic diverted from Wikipedia, the amount of Wikimedia material reproduced by chatbots, or the economic value of any resulting loss.
Updat3: Updat3 cannot conclude anything about this from the current sources.

Timeline

Developments from this story's sources, dated background we compiled and fact-checked, and our earlier coverage. Each row says which.
  1. Congress passed the Copyright Act, whose Section 107 set out four factors for assessing fair use.
    Background · compiled and fact-checked by Updat3
  2. Wikipedia launched as a free online encyclopedia.
    Background · compiled and fact-checked by Updat3
  3. The Wikimedia Foundation was established to support Wikipedia and other free-knowledge projects.
    Background · compiled and fact-checked by Updat3
  4. The Second Circuit ruled that Google’s searchable book-scanning project was fair use in Authors Guild v. Google.
    Background · compiled and fact-checked by Updat3
  5. The immediate legal backdrop was *The New York Times Co. v. Microsoft Corp. and OpenAI*, filed in federal court on December 27, 2023.
    Background · from our historical context
  6. A federal judge ruled that Ross Intelligence’s use of Thomson Reuters legal materials to develop an AI legal research tool was not fair use.
    Background · compiled and fact-checked by Updat3
  7. A federal judge ruled that Anthropic’s use of books to train its AI model was fair use in Bartz v. Anthropic.
    Background · compiled and fact-checked by Updat3

How we got here

The immediate legal backdrop was *The New York Times Co. v. Microsoft Corp. and OpenAI*, filed in federal court on December 27, 2023. The Times alleged that the companies used its articles without permission to train AI systems and that those systems could produce material resembling its journalism; Microsoft and OpenAI disputed the claims.

The governing US framework is the Copyright Act of 1976, whose Section 107 lists four factors for assessing fair use, including the purpose of a use and its effect on a work’s market. In *Authors Guild v.

Original sources

5 articles from 5 outlets

Every story links to the reporting it was built from. How we choose sources

Spotted a factual error, a broken link or a misleading summary? Tell us.