Unsealed filings show OpenAI and Microsoft privately warned of an AI 'doom loop' for the web

Newly unsealed documents enter the record

Court filings unsealed in the New York Times' copyright lawsuit against OpenAI and Microsoft contain internal statements in which the companies' own staff warned that large-scale data harvesting would destabilise the economics of the open web, according to The Verge. The 92-page filing, assembled by the Times' legal team, quotes internal documents and testimony from figures including Microsoft CEO Satya Nadella, OpenAI co-founder Greg Brockman, OpenAI Policy Director Jack Clark, and Nick Turley, OpenAI's head of ChatGPT.

The 'doom loop' warning

Among the most striking material is an internal Microsoft document stating that the company's AI content strategy had started a "doom loop" that would hurt both model performance and the web itself. The document observes that it is highly unusual for an end product to threaten the economic foundations of its essential suppliers, and describes that as exactly the situation Microsoft had created for its LLM business and its "content supply chain." Elsewhere, Microsoft is quoted conceding that LLMs are "a product that destroys its own supply chain" because the output substitutes for the very content used to train it.

Nadella, per The Verge, acknowledged that chatbots have largely replaced search for many people, removing the need to visit original sources. Turley is quoted saying that once a chatbot provides an answer, there is "no good reason to click" through to the publisher. OpenAI's own media and economic experts reportedly attributed falling referral traffic for publishers to AI summaries such as Google's AI Overviews, and speculated that search referrals may be down by as much as 60 percent.

Paywalls, regurgitation and 'the largest theft'

The filing also addresses how training data was assembled. Although Nadella was quoted saying that "anything that is paywalled should be licensed," an OpenAI representative admitted being "unaware" of any effort to detect or remove paywalled content from training data.

Internal OpenAI discussions cited by The Verge show staff understood that GPT-4 had "memorized a ton of data" and would be "insanely good at regurgitation," even while acknowledging that preventing memorisation mattered for limiting copyright violations. The filing includes examples of ChatGPT reproducing long passages from articles in the New York Times, Mercury News, The Denver Post, Lifehacker and Eurogamer.

Microsoft's Director of Applied Science, Brent Hecht, is quoted calling the harvesting behind ChatGPT and Copilot the "largest theft of labor in human history" and saying that Microsoft's legal defence makes a "complete mockery" of fair use. Clark, meanwhile, described building "systems that substitute for the labor" of the people who define society's culture, and internal documents reportedly characterised ChatGPT as "the modern newsstand." Brockman is cited as being focused on the "gazillions" of dollars commercial AI could generate.

Microsoft distances itself

Microsoft has moved to contain the damage. Spokesperson Alex Haurek told The Verge that Hecht's comments reflect one employee's individual perspective, are not a legal analysis, and do not represent the company's views. In a separate filing, Jordan Usdan, GM for Data Strategy and Ops at Microsoft AI, described Hecht's role as bringing "asymmetrical, futuristic, and academic points of view" rather than speaking for the company. Haurek also argued that Nadella's testimony about broad shifts in how people find information is consistent with Microsoft's legal position on the copyright questions before the court.

Why it matters

The unsealed material cuts in two directions. Legally, internal acknowledgements that paywalled content was not filtered out, that models memorise copyrighted text, and that at least one senior researcher viewed the practice as mass uncompensated taking could complicate the fair use arguments OpenAI and Microsoft have built their defence on. Industry-wide, the "doom loop" framing is no longer an outside critic's prediction but an internal assessment: AI products that divert traffic away from publishers also erode the incentive to produce the open web content future models need. As The Verge notes, the documents suggest both companies foresaw harm to publishers and to their own supply chain, and proceeded regardless. The outcome of the Times case may help determine who ultimately pays for that trade-off.