The most damaging words in a lawsuit are often not the plaintiff's. They are the defendant's own, written before anyone thought a court would read them. That is the shape of the documents unsealed this week in the long-running copyright case brought by The New York Times and a dozen other publishers against OpenAI and Microsoft. In internal memos and messages, employees at both companies described what they were building in terms far harsher than anything their lawyers would ever say in public.
One Microsoft director of applied science called the underlying practice "an astonishing theft of unprecedented proportions," and, in a phrase now quoted everywhere, possibly "the largest theft of labor in human history." Nick Turley, an executive who helped lead ChatGPT, wrote that AI posed an "existential threat" to publishers, whose work he described as "largely substitutive" and getting more so. Another memo framed the whole arrangement plainly: "It is highly unusual that an end-product threatens the economic foundations of its essential suppliers, but that is the situation we have created."
The filings put numbers to the scale. The publishers say OpenAI drew on more than ten million articles, close to a third of them from the Times, and that its models trained on a corpus whose licence explicitly limited use to non-commercial research. They also surface the tone of the people involved. Co-founder Greg Brockman is quoted admitting he was "deeply motivated by the gazillions" he hoped the technology would generate. When a researcher reported finding a way around the Times paywall, the response was reportedly "ahead nice."
What makes the memos land is the gap between them and the companies' legal posture. In court, OpenAI and Microsoft argue that training on copyrighted material is transformative and protected by fair use, a position the US Department of Justice has backed in a brief invoking scientific progress and national security. Microsoft has distanced itself from its own employee's language, calling the "theft" remark "one employee's individual perspective" that does not represent the company's views. But Satya Nadella conceded under oath that a chatbot answering a question "has substituted" the trip to the original website, which is precisely the harm the publishers allege.
That harm is not hypothetical, and it is not confined to big newspapers. Independent outlets describe a sharper squeeze from AI summaries that answer readers before they ever click through. Truthout, one of the publications covering the filing, says it lost roughly 15 percent of its readership almost overnight as AI-generated overviews rose to the top of search results, and expects to forgo around half a million dollars in online revenue over the coming year. Even OpenAI's own engineers seem resigned to the dynamic. "No matter how prominently we show the links, users won't click," one is quoted saying.
The legal stakes are large but slow. The publishers have asked for summary judgment; if the judge grants it the case skips trial, and a ruling is not expected before 2027. The deeper tension is the one the memos name so precisely. These systems are good at news because they learned from news, and the better they get, the less traffic and revenue flows back to the people who produce it. A memo inside Microsoft had a word for that loop. It called it a "doom loop," one that would eventually degrade the models along with the web they feed on. The companies saw it clearly. They built it anyway.