⚖️ New York Times v OpenAI: 'billions' sought, fair use claimed

The New York Times sued OpenAI and Microsoft on Dec. 27, 2023, in Manhattan federal court, accusing them of using millions of its articles without permission to help train chatbots, Reuters reported. The case is New York Times Co v Microsoft Corp et al, No. 23-11195. The Times put no number on its claim. It estimated damages in the 'billions of dollars'.

The relief it wants is blunt. Per Reuters, the Times asks the court to make OpenAI and Microsoft destroy chatbot models and training sets that incorporate its material. The Verge reported that it also asks the court to stop the companies from training on its content.

The money on the table

Reuters put the stakes in context at filing: investors valued OpenAI at more than $80 billion, and Microsoft had invested $13 billion in a for-profit subsidiary, for what would be a 49% stake. The Times, Reuters noted, ended September 2023 with 9.41 million digital-only subscribers, and subscriptions generate more than two-thirds of its revenue. One side is a valuation. The other is a subscription base. Both are claims about the future.

"There is nothing 'transformative' about using The Times's content without payment to create products that substitute for The Times and steal audiences away from it,"

The New York Times, in its complaint, as quoted by Reuters, Dec. 27, 2023

The defense

OpenAI and Microsoft have said that using copyrighted works to train AI amounts to fair use, Reuters reported. On the day of filing, OpenAI said in an emailed statement that its conversations with the Times had been 'productive and moving forward constructively', and that it was 'surprised and disappointed'. Reuters said Microsoft did not respond to its requests for comment.

By February 2024 the tone had hardened. The East Bay Times reported that the two companies, responding to the Times, called the claim that generative AI threatens journalism 'pure fiction', and accused the paper of using 'deceptive' prompts to produce 'highly anomalous results'. Reuters reported that OpenAI separately accused the Times of 'hacking' its systems. The Times replied that it was 'simply using OpenAI's products to look for evidence' of reproduction of its copyrighted work.

What the record shows about training data

The complaint's evidence is outputs. Reuters reported that it cited chatbot answers containing near-verbatim excerpts, including a Pulitzer Prize-winning 2019 series on predatory lending in New York City's taxi industry and a 2012 restaurant review by Pete Wells. The Guardian reported that the suit also flagged a Bing Chat answer on 'the 15 most heart-healthy foods', where, the lawsuit claims, twelve items were not in the Times story it was credited to.

The training sets themselves are harder to see. In a separate suit by The Intercept, the complaint says the defendants 'have kept secret the specific content used to train all versions of ChatGPT beginning with GPT-4'. That is an allegation, filed Feb. 28, 2024, and not a finding.

"it would be impossible to train today's leading AI models without using copyrighted materials"

OpenAI, in a January 2024 filing to the U.K. House of Lords inquiry, as quoted by CNBC

Other cases have produced documents. Business Insider reported that unsealed filings in the Authors Guild class action show OpenAI deleted two datasets, 'books1' and 'books2'. Per OpenAI's 2020 white paper they made up 16% of GPT-3's training data, at 67 billion tokens, and the Guild's lawyers said they probably held 'more than 100,000 published books'. OpenAI's lawyers said the sets were deleted in mid-2022 because of non-use. OpenAI said the models powering ChatGPT today were not developed using them. That is the authors' case, not the Times's.

The Times has also reported on the subject itself. Its April 2024 investigation, citing people with knowledge of the matter, said an OpenAI team transcribed more than one million hours of YouTube videos in late 2021, and that president Greg Brockman personally helped collect them, according to two of the people. OpenAI told the Times that each of its models 'has a unique data set that we curate'. This is the paper's reporting about a rival. It is not a court finding.

The court record

The Times case is no longer standing alone. The New York Times reported in April 2025 that a U.S. judicial panel had consolidated several claims against OpenAI, including the Times's. The Intercept reported on Sept. 2, 2026 that the cases sit before U.S. District Judge Sidney Stein, alongside Tribune Media Company and Ziff Davis.

That filing is the newest document in the pack. The Intercept reported that the Justice Department told the court the U.S. has a 'strong interest' in a competitive AI industry, and backed OpenAI's fair-use defense. The Intercept's attorney, Matt Topic, said the department had not been privy to any discovery in the case and that its position reflects what 'Microsoft's and OpenAI's PR machines have been putting out'. The documents read contain no ruling on discovery in the Times case, so none is reported here.

One more entry for the file. In a separate authors' complaint, reported by Ars Technica, the plaintiffs say Sam Altman told Congress that 'creators deserve control over how their creations are used', and allege that he and OpenAI have 'proved unwilling to turn these words into actions'. The authors say it. OpenAI told Ars it respects the rights of writers. You can weigh the two.