Newly unsealed court documents in the New York Times' copyright case against OpenAI and Microsoft show both companies were internally warning, in their own words, that their AI products would gut the web that trained them. The 92-page filing, unsealed September 18, 2026, quotes an internal Microsoft document describing a 'doom loop' for the model business and its 'content supply chain,' and includes an OpenAI estimate that search referrals to publishers may be down as much as 60 percent because of AI summaries.
The filing pulls from communications by Satya Nadella, Sam Altman, OpenAI cofounder Greg Brockman, OpenAI Policy Director Jack Clark, and the head of ChatGPT, identified as Nick Turley. It also leans heavily on Brent Hecht, Microsoft's Director of Applied Science, who characterized the company's scraping practices as the 'largest theft of labor in human history' and said Microsoft's fair-use defense made 'a complete mockery of the idea of fair use.'
Microsoft has moved to distance itself from Hecht. Spokesperson Alex Haurek said the comments reflect one employee's view and are not a legal analysis. In a separate filing, Jordan Usdan, GM for Data Strategy and Ops at Microsoft AI, said Hecht 'holds divergent, academic, and forward-looking views' and is employed to bring 'asymmetrical, futuristic, and academic points of view.' The framing is that Hecht is a paid in-house contrarian, not a company voice.
“largest theft of labor in human history”— Brent Hecht, Microsoft Director of Applied Science
Key facts
- 01A 92-page filing unsealed September 18, 2026 in the New York Times case cites internal Microsoft and OpenAI documents warning of a 'doom loop' for the open web.
- 02OpenAI's own media and economic experts attributed as much as a 60 percent drop in search referrals to AI summaries like Google AI Overviews.
- 03Microsoft Director of Applied Science Brent Hecht called AI training scraping the 'largest theft of labor in human history' — a phrasing the company has publicly disavowed.
- 04OpenAI admitted GPT-4 'memorized a ton of data' and the filing shows verbatim regurgitation of articles from the Times, Mercury News, Denver Post, LifeHacker, and Eurogamer.
- 05An OpenAI representative said he was 'unaware' of any effort to detect or remove paywalled content from training data, despite Satya Nadella saying paywalled content 'should be licensed.'
The most quotable line from Microsoft's own documents is the doom-loop passage itself: '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.' A separate internal note bluntly states that 'LLMs are a product that destroys its own supply chain.'
On paywalls, the filing surfaces a gap between Nadella's stated principle and OpenAI's practice. Nadella is quoted saying 'anything that is paywalled should be licensed.' An OpenAI representative, according to the Times, said he was 'unaware' of any effort to detect or remove paywalled content from training data.
“These comments reflect one employee's individual perspective, are not a legal analysis, and do not represent the company's views.”— Alex Haurek, Microsoft spokesperson
The filing also documents what OpenAI knew about memorization. Employees acknowledged that preventing memorization was important to 'minimize copyright violations,' while also noting that GPT-4 'memorized a ton of data and therefore will be insanely good at regurgitation.' The Times filing then reproduces examples of ChatGPT outputting long strings of copy from the Times, Mercury News, The Denver Post, LifeHacker, and Eurogamer in response to queries.
On the business rationale, Brockman is quoted expressing interest in the 'gazillions' of dollars available in commercial AI. Clark, on the labor question, wrote that the company was 'creating systems that substitute for the labor of the people that define the culture of society.' Internal OpenAI documents describe ChatGPT as 'the modern newsstand,' and Turley is quoted saying that once a user gets an answer from the chatbot, there is 'no good reason to click' on the source.
“creating systems that substitute for the labor of the people that define the 'culture' of society”— Jack Clark, OpenAI Policy Director
The referral-traffic collapse is now measurable rather than theoretical. OpenAI's own media and economic experts attributed publisher declines directly to AI summaries, including Google AI Overviews, and estimated the drop at as much as 60 percent. Publishers have been describing this dynamic as 'Google Zero' for two years; the filing shows the AI companies were modeling it internally at the same time.
Microsoft's counter is narrower than the quotes suggest. Haurek said Nadella's testimony and Microsoft's position in the case are 'perfectly consistent,' arguing that Nadella was describing broad changes in how people find information, not conceding copyright questions before the court. Microsoft and OpenAI have not yet filed their full response to the newly unsealed exhibits, and the material shown so far is the Times' selection, not a full record. A memorization example in a demo prompt is also not the same as a finding of infringement at trial.
Still, the filing changes the shape of the case. Publishers have spent two years arguing that generative AI is a substitutional product built on their work; the unsealed material shows that inside Microsoft and OpenAI, senior people were writing the same sentences. That is useful evidence for the Times on both willfulness and market harm, the two levers that decide statutory damages and the fair-use analysis.
For the AI market, the immediate consequence is licensing leverage. Every publisher currently negotiating with OpenAI, Microsoft, Google, and Anthropic now has a filing on the docket in which the defendants' own employees describe the scraping as theft and the downstream traffic loss at up to 60 percent. Deal prices go up, holdout publishers get bolder, and the cost basis of frontier training data — long treated as effectively zero — starts to look like a real line item. The doom-loop memo was written to warn Microsoft's leadership; it will now be read by every plaintiff's lawyer in the queue.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
