The Times Just Quoted OpenAI And Microsoft Against Themselves. A Microsoft Director Called It The “Largest Theft Of Labor In Human History.”
The news plaintiffs’ summary judgment brief is built on the defendants’ own documents. Here is what it says their people wrote, and the numbers behind it.
The New York Times Building in Manhattan, 2011. Photo: Eden, Janine and Jim, Creative Commons Attribution 2.0, via Wikimedia Commons.
Most AI lawsuits are fought with the plaintiff’s words. This one is being fought with the defendants’ own.
On September 17, 2026, The New York Times and the other news plaintiffs suing OpenAI and Microsoft filed their combined summary judgment brief in federal court in Manhattan. It runs 92 pages, and a lot of the best lines in it were written by people who work at the two companies being sued. TechCrunch was first to report the unredacted version that afternoon.
Here’s the opening of the brief, word for word:
“This case is about, as Microsoft’s Director of Applied Science put it, ‘an astonishing theft of unprecedented proportions’; perhaps the ‘largest theft of labor in human history.’”
That’s not a newspaper describing itself as a victim. That’s the newspaper quoting a Microsoft director, Brent Hecht, according to the filing and the reporting on it.
One caution up front, because it matters. This is a brief. It’s one side’s argument, built on documents picked to win. Nothing here is a ruling, and OpenAI and Microsoft didn’t return TechCrunch’s requests for comment. But the quotes are attributed to their own people, and each one carries a citation to the evidence record.
What their own people wrote, according to the brief
OpenAI’s Head of ChatGPT wrote that publishers face an “existential threat” from these products, which he said “are largely substitutive, period” and “will get more and more substitutive as they get better.” TechCrunch identifies him as Nick Turley. The brief says OpenAI understood that threat as early as June 2023.
The same Microsoft executive, the brief says, recognized that a win for the defendants on fair use would arguably “make a complete mockery of the idea of ‘fair use.’”
An OpenAI software engineer, per the brief: “no matter how prominently we show the links, users won’t click.”
And a Microsoft document the brief quotes at length:
“Our AI content strategy has started a ‘doom loop’ that will hurt the performance of our models and the entire web at the same time.”
The same passage goes on to say it’s highly unusual for an end product to threaten the economic foundations of its essential suppliers, “but that is the situation we have created for our LLM business with respect to its ‘content supply chain.’”
“ah nice”
This is the line that’ll get quoted for years. According to the brief, when OpenAI’s Nick Ryder told cofounder Greg Brockman about “a hack to get around nytimes paywall” to help Brockman’s effort to scrape the Times’s site, Brockman replied:
“ah nice.”
The brief also says OpenAI’s corporate representative wasn’t aware of any method at the company for detecting paywalled content in its datasets. And it points to Custom GPTs hosted on OpenAI’s own store with names like “Remove Paywall” and “Article Reader,” which advertised summarizing “paid (paywalled items). Just submit the URL.”
Here’s why that lands so hard. Microsoft CEO Satya Nadella testified, per the brief, that “anything that is paywalled should be licensed by anyone who wants to use it” for grounding or training. He also said that if he’d been told OpenAI scraped and trained on paywalled material, he would have invoked Microsoft’s right to require OpenAI to retrain its models.
So Microsoft’s CEO says paywalled work has to be licensed, and the brief says OpenAI’s cofounder said “ah nice” to a paywall hack. That’s the gap the Times wants a judge to look at.
The numbers in the filing
- 91,692 copies of the plaintiffs’ works in OpenAI’s mid training datasets, according to the brief.
- 160,903 unique plaintiff works in a training set assembled from Microsoft’s Project Mango data.
- 6,552 scraped Times works, 18,609 from the Daily News publishers and 66,780 from Ziff Davis in WebText2, an expanded version of an early OpenAI training set.
- More than 1.8 million articles in The New York Times Annotated Corpus, which the brief says OpenAI obtained from a third party. It covers nearly every Times article from January 1, 1987 through June 19, 2007.
- 87 to 93 percent lower click through rates for the Times on Copilot than on traditional Bing search, according to Microsoft’s own data as described in the brief. For the Daily News publishers it was 83 to 91 percent. For Ziff Davis, 51 to 94 percent.
That last one is the doom loop in a single line. The brief says Copilot’s home page greeted users with: “Instead of clicking through links, we can talk through whatever you’re curious about.” Microsoft then measured what happens to the sites nobody clicks through to.
Why this one is different
AI companies usually defend training with a clean story. The model learned from the open web the way a person learns from reading, and the output is something new.
This brief goes after every piece of that story with internal documents. It asks for summary judgment on liability at each stage: getting the articles, often from behind paywalls, training on them, grounding answers in them through Bing, and putting copies back out. It argues the products are substitutive, which the brief calls “copyright’s bête noire,” quoting the Supreme Court’s 2023 decision in Andy Warhol Foundation v. Goldsmith.
If the court agrees the products substitute for the journalism they were built on, the fair use defense gets very thin. That’s why the internal quotes matter so much. The word “substitutive” didn’t come from the Times. According to the brief, it came from OpenAI.
What happens next
The case is In re OpenAI, Inc. Copyright Infringement Litigation, consolidated before Judge Sidney H. Stein in the Southern District of New York. The Times first sued in December 2023. The defendants will answer this brief, and the judge decides whether any of it can be resolved without a trial.
We’ll keep tracking it on our lawsuit tracker. For now, the most damaging thing anyone has said about how these models were built might be two words from one of the people who built them.
Sources: News Plaintiffs’ Combined Summary Judgment Brief, Document 1977 1, filed September 17, 2026, via CourtListener; TechCrunch, September 17, 2026; The Washington Post, September 17, 2026; Engadget, September 18, 2026. Photo: New York Times Building, Eden, Janine and Jim, 2011, Creative Commons Attribution 2.0, via Wikimedia Commons.
