Newly unsealed court filings in the New York Times' landmark copyright lawsuit against OpenAI and Microsoft show the companies' own executives warned internally that scraping news content for AI training would devastate publishers, even as both firms pressed ahead with the practice.
The documents, made public Friday in a Lower Manhattan courthouse, landed with force in the ongoing legal battle over whether Silicon Valley's largest AI companies built their products on stolen work. Internal communications from senior figures at both OpenAI and Microsoft now form part of the court record, and they paint a picture sharply at odds with the defendants' public posture that training AI on copyrighted material qualifies as fair use under federal law.
Among the most damaging revelations: Brent Hecht, Microsoft's Director of Applied Science, described the AI training regime in internal communications as "the largest theft of labor in human history," the New York Post reported. Hecht went further, warning that Microsoft's AI content strategy had created what he called a "doom loop" threatening the very publishers whose work fed the models.
"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. 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."
That was Hecht's own internal assessment, not the argument of a plaintiff's attorney. It came from inside the company now arguing in court that its use of copyrighted journalism was perfectly legal.
OpenAI's internal record is no less revealing. Greg Brockman, the company's co-founder, allegedly responded "ah nice" when an employee informed him of a hack that bypassed the New York Times' paywall, a system the newspaper uses to fund the journalism OpenAI was copying. The Financial Times first reported the exchange, which appears in the unsealed filings. Brockman also wrote in 2017 that he was "deeply motivated by the gazillions" he hoped to earn by commercializing AI technology, CNBC reported.
A separate unnamed OpenAI executive allegedly acknowledged that the company's chatbots posed an "existential threat" to journalism. The Wall Street Journal attributed that admission to the unsealed documents. The New York Post identified the executive as Nick Turley, OpenAI's ChatGPT chief, and reported that Turley acknowledged AI products would become increasingly "substitutive", meaning they would replace, not supplement, the original news content they were trained on.
So the people building these tools understood the damage. They said so in writing. And they kept going.
The lawsuit has drawn interest well beyond the publishing industry. On September 1st, the U.S. Department of Justice filed what is known as a statement of interest, a formal brief asking the court to rule that training AI models on copyrighted work qualifies as fair use. The DOJ's rationale centered on national security: if foreign adversaries will train their AI systems on anything they can access by any means, restricting American companies risks handing a permanent technological advantage to rivals like China.
That argument may carry weight in a courtroom, but it puts the federal government squarely on the side of the companies whose own executives described their conduct as theft. The DOJ did not address the internal admissions in its filing.
For the New York Times, the unsealed documents bolster a straightforward claim: OpenAI copied millions of copyrighted articles to train its models, exploited technical workarounds to bypass the newspaper's paywall, and the companies' own data showed that Times click-through rates dropped precipitously as a result. The newspaper is seeking a pre-trial ruling in its favor.
NYT shares barely moved on Friday, and the stock remains flat year-to-date. But options traders saw something worth betting on. Roughly 4,400 October 72.5/77.5 call spreads traded at a $0.925 debit, risking just 1.3 percent of Friday's closing price. That trade pays off with a gain of at least 4.6 percent by October expiration, and could return 10 percent or more if the stock reaches the $77.73 average analyst target.
At the same time, volume picked up in October 62.5 and 65 puts, suggesting some traders are hedging the downside. A combined 65/77.5 call spread offers more than $7 of upside against roughly $5 of downside, a favorable risk-reward profile heading into what could be a decisive phase of the litigation.
The options activity reflects a market view that a summary judgment ruling or settlement could arrive within weeks. If the Times prevails, the financial implications extend far beyond one newspaper's stock price. Every publisher whose work was fed into AI training models without permission would have a precedent to cite.
The unsealed record includes more than rank-and-file warnings. Microsoft CEO Satya Nadella testified that paywalled content should require licensing, an admission that undercuts his own company's legal defense. If the head of Microsoft believes licensing is appropriate, the fair-use argument starts to look less like a principled legal position and more like a post-hoc rationalization for conduct the company's leadership knew was problematic.
The case now consolidates suits from the Times and from newspapers owned by Alden Global Capital, broadening the class of publishers seeking accountability. The question before the court is not abstract. It is whether the largest technology companies in the world can take the work of journalists, use it to build products worth billions, and owe nothing to the people who produced it.
The defendants' own words suggest they already know the answer. When your director of applied science calls your business model "the largest theft of labor in human history" and your co-founder celebrates a paywall hack, the fair-use defense starts to read less like law and more like wishful thinking.
Copyright exists to protect the people who create things from the people who take them. If the courts won't enforce it here, it's hard to see where it means anything at all.