NYT v. Microsoft/OpenAI: 'Largest Theft of Labor'

Update, Sept. 18

Update, Sept. 18: The Wall Street Journal's Sept. 17 reporting on the same unsealed filing adds three details this article didn't include. Hecht's "doom loop" memo was written "soon after the New York Times sued," per the Journal — the warning came early, not in hindsight. A Microsoft researcher told colleagues that compensating content creators is "in the best interest of my employer, of my country, and of many other groups I belong to." And OpenAI employees discussed "opportunities to skirt news sites' paywalls" — plural opportunities, beyond the single NYT paywall workaround already documented here. Separately, 404 Media's Sept. 17 account of the filing notes an OpenAI software engineer testified that "no matter how prominently we show the links, users won't click" — the substitution problem stated as bluntly as the 93% figure above. No Sept. 18 developments in the case have surfaced as of this writing.

The most damaging evidence in The New York Times' copyright case against OpenAI and Microsoft didn't come from the Times' lawyers. It came from the defendants' own offices. In materials unsealed Thursday in the Southern District of New York — nearly three years after the Times filed suit on Dec. 27, 2023 — Microsoft's director of applied science, Brent Hecht, is quoted calling the mass scraping of news articles "an astonishing theft of unprecedented proportions" and, in a January 2023 memo, "the largest theft of labor in human history."

Before going further, a necessary caution: the underlying exhibits remain sealed. The new quotes come from the Times' own unredacted brief — an adversarial source — and, as TechCrunch noted in its reporting, they are presented without their original context. What follows is what the Times says the documents show, not what a judge has found.

With that caveat, the picture the Times paints is stark. Its motion for summary judgment argues that Microsoft and OpenAI built their businesses on copying millions of news articles without permission, bypassed the Times' paywall to do it, and privately acknowledged that their products threaten the economic foundations of the publishers whose work trained them — even as both companies have argued in court that training AI on copyrighted material is protected fair use.

The New York Times Building at 620 Eighth Avenue in Manhattan, photographed in July 2019. The Times and other publishers say newly unsealed filings in their copyright suit show Microsoft and OpenAI knew their AI products threatened the economics of news. Photo: Ajay Suresh, via Wikimedia Commons (CC BY).

The memos that call it theft

Hecht's memos, as quoted in the filing, run directly against the fair-use defense Microsoft and OpenAI have mounted. If a company's own AI training is fair use, the Times argues, then according to Hecht it would "make a complete mockery of the idea of 'fair use.'" In a January 2024 internal presentation, Hecht described the dynamic between AI answer engines and publishers as a "doom loop" that would "hurt the performance of our models and the entire web at the same time."

Another Microsoft document put the structural problem more starkly: "It is highly unusual that an end-product threatens 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.'"

Microsoft, through a spokesperson, pushed back. The company said Hecht's comments "reflect one employee's individual perspective, are not a legal analysis, and do not represent the company's views." It argues that AI training is transformative and therefore protected by fair use, and that its Copilot chatbot doesn't substitute for publishers' journalism. That defense will have to survive the next piece of the filing.

Microsoft's own data: the 93% click-through collapse

Fair use turns heavily on market harm: a use is less likely to be fair if it substitutes for the original work. The Times says Microsoft's own data makes that case for it. According to the filing, the company's Copilot "answer engine" caused click-through rates for The New York Times' domain to drop as much as 93% compared with traditional Bing search — users getting answers from the AI instead of visiting the paper's site.

That figure is the filing's hardest empirical punch. It's one thing for a publisher to argue that AI summaries might reduce traffic; it's another to have the defendant company's own measurements showing a near-total collapse in referral clicks. Microsoft's defense — that Copilot doesn't substitute for journalism — sits uncomfortably beside a 93% decline in the mechanism by which readers reach journalism.

How OpenAI got around the NYT paywall

The filing also reconstructs how the content was allegedly obtained. When OpenAI researcher Nick Ryder told company president Greg Brockman about a "hack to get around nytimes paywall" to aid scraping efforts, Brockman replied: "ah nice."

The exchange is brief, but the Times clearly included it for maximum effect: it suggests evasion of the paper's paywall wasn't an accident of web crawling but a deliberately celebrated workaround. The filing further alleges that OpenAI and Microsoft scraped content from the Bing index, and that researchers deliberately stripped copyright notices from training data — one reason given being that they "wouldn't want model outputting" copyright notices to users.

"Existential threat"

OpenAI's own leadership, as quoted in the brief, acknowledged the substitution problem. Nick Turley, the company's head of ChatGPT, wrote in an internal communication that publishers face an "existential threat" from products like the chatbot, which are "largely substitutive" and "will get more and more substitutive as they get better."

Brockman described the models as "excellent at news." Microsoft CEO Satya Nadella, testifying in a deposition earlier this year, said that "anything that is paywalled should be licensed by anyone who wants to use it… for grounding or training" — and that, had he known OpenAI had scraped paywalled information, he would have "invoked [Microsoft's right to] require OpenAI to retrain its models." Under oath, Nadella also agreed that conversing with chatbots "has substituted… giving you the information right there on the website on the AI platform versus needing to go to the underlying source."

A Microsoft document added a further wrinkle, warning of a "real risk" that generative AI could "significantly disrupt the employment of the very people who generated the data on which the foundation model was trained." OpenAI policy director Jack Clark wrote that the company's work "is going to increasingly lead to us creating systems that substitute for the labor of the people that define the 'culture' of society."

The defendants also allegedly tried to contain the evidence trail. The Times says that immediately after the publishers filed suit, OpenAI built a filter to suppress output of their content — but not content from publishers who hadn't sued. Hecht reportedly called that behavior an "accidental cover up," because it would leave "people who have a right over the content having less visibility into what was used for training." (On OpenAI's handling of internal misconduct, see also: the rogue agents that probed Hugging Face for months before their July disclosure.)

The scale of the copying

The filings put numbers to the ingestion for the first time. OpenAI's mid-training datasets alone contained more than 91,692 copies of works published by the Times, the Daily News, and the Center for Investigative Reporting. A dataset derived from Common Crawl included more than 2 million documents from nytimes.com alone.

The companies also allegedly built training-data pipelines with their own project names: "OpenAI delivered the entire GPT-3 training dataset to Microsoft," the filing reads, while Microsoft supplied training data to OpenAI through initiatives called Project Taxi and Project Mango. The Project Mango dataset, the filing says, contains copies of at least 160,903 unique works from the news publishers. OpenAI employees also allegedly built training datasets, WebText and WebText2, that disproportionately relied on scraped news content, and pulled millions of articles from Common Crawl.

OpenAI and Microsoft did not return requests for comment on the filings.

Why the Times' week wasn't all good news

Analysis — the following is interpretation, not reported fact. For all the damaging quotes, the Times took a meaningful loss this week too. Bloomberg Law reported that Judge Stein rejected the paper's bid to add inducement-based contributory-infringement claims — the judge ruled the request came too late. That matters because contributory claims would have widened the circle of liability; without them, the Times must win on its existing theories, primarily direct infringement.

The broader legal landscape still favors the AI companies. Judges have been largely receptive to fair-use arguments for AI training, and earlier this month the Trump administration filed a brief defending the unlicensed use of copyrighted material to train large language models — a significant thumb on the scale. A separate $1.5 billion Anthropic copyright settlement, approved in July, covers different claims against a different company and doesn't set the rules for this fight.

So why does the unsealing matter? Because summary judgment — the motion this material was unsealed in — is where cases are often won before trial. If the Times can show there is no genuine dispute that Microsoft and OpenAI knew their products would substitute for the publishers' work and built them anyway, the fair-use defense gets much harder. Internal documents describing your own business as a "doom loop" for your suppliers are the kind of evidence that moves judges. The companies will argue the quotes are cherry-picked, taken out of context, and in some cases from employees speaking personally rather than for the company — exactly the argument Microsoft's spokesperson already made about Hecht.

The honest assessment is that neither side got a knockout. The Times now has the defendants' own words on substitution and market harm; the defendants still hold the fair-use precedents and the administration's support. What the unsealing guarantees is that the next phase of this three-year fight will be conducted in public — with the most quoted internal memos in AI copyright law now part of the permanent record. And it lands in the middle of the wider AI-safety fight that has split Silicon Valley and Washington — a fight in which who controls the training data is becoming the central question.

NYT v. Microsoft/OpenAI lawsuit FAQ

What is the NYT's lawsuit about? The Times sued Microsoft and OpenAI on Dec. 27, 2023, alleging they copied millions of its articles without permission to train AI models; the companies argue training is protected fair use.

What is summary judgment? The unsealed material comes from the Times' motion for summary judgment — a request that the judge rule in its favor without a trial. It is where cases like this are often won or lost before trial.

What did the companies' own people say? Microsoft's Brent Hecht called scraping "the largest theft of labor in human history" and described a "doom loop"; OpenAI's Nick Turley called chatbots an "existential threat" to publishers; CEO Satya Nadella testified that "anything that is paywalled should be licensed."

What went against the Times this week? Judge Stein rejected the paper's bid to add inducement-based contributory-infringement claims, ruling the request came too late.

Sources: TechCrunch, "Microsoft exec called AI scraping 'the largest theft of labor in human history,' new unredacted filings reveal" (Sept. 17, 2026); Bloomberg Law, "OpenAI, Microsoft Leaders Admitted AI Threatens News Outlets" (Sept. 17, 2026); Reuters, "New York Times sues OpenAI, Microsoft for copyright infringement" (Dec. 27, 2023).