Worn, unlabelled bound volumes lean on tall grey steel shelves along both sides of a narrow archive room, lit by a single hanging lamp.

Copyright

OpenAI Defends Its Training Against the Times and Authors

OpenAI and Microsoft told a judge that training on news and books was fair use; OpenAI says its expert found 24 verbatim reproductions in 20 million ChatGPT conversations.

NEW YORK, Oct. 23 — OpenAI and Microsoft told a federal judge in Manhattan on Friday that using newspaper articles and books to train AI models was fair use, and that the publishers and authors suing them had suffered no harm the law recognizes. “In short, an LLM is not a book machine. It is an everything machine,” Microsoft wrote.

Their briefs, made public with redactions, oppose the motions for partial summary judgment that The New York Times and other news publishers, and a group of authors, filed on Sept. 4. Both sides want Judge Sidney H. Stein to decide fair use without a trial.Fair UseA defense written into the Copyright Act. Courts weigh four factors: the purpose and character of the use, the nature of the work, the amount used, and the use’s effect on the “market for or value of the copyrighted work.”

At stake is whether AI companies may copy these works for training without permission. OpenAI argues that there is “no traditional or likely-to-be-developed market for licensing news for pretraining,” and that the deals the publishers point to “center on materially different rights, such as the right to display verbatim content from partner websites.” In their own brief, filed Friday with redactions, the publishers answered that “OpenAI and Microsoft have themselves already licensed media content to train their GAI models.” Amazon, Google, Meta, Perplexity, ProRata.ai and Mistral AI have licensed news for training too, they wrote, evidence of “a developing market.”

OpenAI’s defense against the Times turns on a split. The publishers argue that ChatGPT keeps readers from their websites by giving “fast, timely answers.” That ability, OpenAI wrote, “arises not from pretraining, but from a distinct ‘web browsing’ feature, which News Plaintiffs’ motion does not challenge.” In their September motion, the publishers put off seeking summary judgment against OpenAI on what ChatGPT fetches and prints until their sanctions motion is resolved. On Friday they argued that browsing, too, “entails wholesale copying of Plaintiffs’ works” and is not fair use.

On how often ChatGPT repeats the publishers’ words, OpenAI cited the experts. Its own expert “identified only 24 instances of verbatim reproduction in a sample of 20 million ChatGPT conversation logs, a rate of 0.00012%.” By OpenAI’s account, the publishers’ experts found rates from 0.000002 percent, for The Intercept, to 0.00011 percent, for The Times and the Daily News publishers.The ModelsThe case concerns seven OpenAI models, the briefs say: GPT-3, GPT-3.5, GPT-3.5 Turbo, GPT-4, GPT-4 Turbo, GPT-4o and GPT-4o mini. In the section of their brief on the removal of copyright-management information, a separate claim, the publishers answered that OpenAI’s argument that “regurgitation hardly ever happens in the first place” is “both disputed and non-dispositive,” and that regurgitation of their works “is widespread when extrapolated over the billions of outputs from OpenAI’s products.”

On harm, the publishers wrote that “it is Defendants’ burden to show a lack of actual or potential market harm.” The authors’ opposition briefs are sealed; a redacted copy of one, opposing Microsoft, was docketed on Saturday, and this paper has not seen it. In September the authors wrote that “OpenAI did not even buy the books it used,” pointing to downloads from LibGen, which they call a pirate site: “It is no exaggeration to say that OpenAI built the foundations of its business on mass piracy.”

OpenAI says that of the models at issue only GPT-3 and GPT-3.5 were trained on material downloaded from LibGen, and that it stopped using LibGen books in late 2021. The authors also quote a 2025 ruling in Bartz v. Anthropic, a case in California: “Such piracy of otherwise available copies is inherently, irredeemably infringing.” OpenAI calls that statement “explicitly dicta.”

Other papers are waiting on Judge Stein’s ruling. Suits against OpenAI by The Seattle Times and Newsday, filed on Sept. 4, and by USA Today and other newspaper companies, filed on Oct. 8, are stayed until the summary-judgment motions are resolved.

Replies are due Nov. 20 under a schedule the judge approved, and on Friday OpenAI, Microsoft and the publishers asked him to hear oral argument. The authors asked for argument in September. Friends-of-the-court briefs are due Oct. 30.

Daily Cortex is written with AI models, including Anthropic’s Claude, and fact-checked with OpenAI’s GPT models.

Sources

  1. OpenAI (S.D.N.Y. Doc. 2207), “OpenAI’s Opposition to News Plaintiffs’ Motion for Partial Summary Judgment (public version, redacted), In re OpenAI, Inc. Copyright Infringement Litigation, No. 25-md-3143,” Oct. 23, 2026.
  2. OpenAI (S.D.N.Y. Doc. 2203), “OpenAI’s Opposition to Class Plaintiffs’ Motion for Summary Judgment (public version, redacted),” Oct. 23, 2026.
  3. Microsoft Corporation (S.D.N.Y. Doc. 2173), “Defendant Microsoft Corporation’s Opposition to Books Plaintiffs’ Motion for Partial Summary Judgment (redacted, public version),” Oct. 23, 2026.
  4. The New York Times Company et al. (S.D.N.Y. Doc. 2118), “News Plaintiffs’ Omnibus Memorandum of Law in Opposition to Defendants’ Motions for Summary Judgment (redacted),” Oct. 23, 2026.
  5. The New York Times Company et al. (S.D.N.Y. Doc. 1709), “News Plaintiffs’ Combined Summary Judgment Brief (redacted version),” Sept. 4, 2026.
  6. Class Plaintiffs (S.D.N.Y. Doc. 1881), “Class Plaintiffs’ Memorandum of Law in Support of Motion for Partial Summary Judgment (redacted, publicly filed),” Sept. 5, 2026.
  7. U.S. District Court, S.D.N.Y. (Judge Sidney H. Stein, Doc. 2049), “Letter on the summary-judgment schedule, memo endorsed: “The schedule as set forth above is granted.,” Sept. 24, 2026.
  8. U.S. District Court, S.D.N.Y. (Judge Sidney H. Stein, Doc. 2072), “Joint Stipulation and Order Staying Seattle Times Action,” Sept. 28, 2026.
  9. U.S. District Court, S.D.N.Y. (Judge Sidney H. Stein, Doc. 2110), “Joint Stipulation and Order Staying USA Today Action,” Oct. 16, 2026.
  10. News Plaintiffs (S.D.N.Y. Doc. 2139), “Letter motion requesting oral argument on OpenAI’s and Microsoft’s motions for summary judgment,” Oct. 23, 2026.
  11. U.S. District Court, S.D.N.Y. (CourtListener), “Docket, In re OpenAI, Inc. Copyright Infringement Litigation, No. 1:25-md-03143,” Oct. 24, 2026.