Share with your CEO
Unsealed court documents in the New York Times lawsuit against OpenAI and Microsoft reveal that both companies internally acknowledged their AI content strategy would devastate the publishing ecosystem they depended on for training data. Microsoft’s own documents warned of a “doom loop” destroying the web’s economic foundations, a Microsoft director called the scraping “the largest theft of labor in human history,” and OpenAI’s head of ChatGPT admitted publishers faced an “existential threat” from products he described as “largely substitutive, period.” OpenAI’s experts now concede search referrals to publishers may be down as much as 60 percent since AI Overviews launched.
What this means for your business
The documents establish something harder to argue around than external criticism: the companies knew, in writing, before shipping. That distinction matters to every enterprise leader currently negotiating AI vendor contracts, because it reframes the question from “did harm occur” to “was harm foreseeable and accepted.” Organizations licensing OpenAI or Microsoft AI products that depend on web-sourced content pipelines, competitive intelligence feeds, or third-party publisher data are now operating inside a supply chain that Microsoft’s own internal documents described as one that “destroys its supply chain.”
The “doom loop” framing in the Microsoft documents is not rhetorical excess from a single dissenting employee, whatever Microsoft’s legal team would prefer the court to believe. The document is institutional, not individual, and Satya Nadella confirmed under oath that chatbots now substitute for clicking through to source content. That combination, an acknowledged substitution effect plus a collapsing referral economy (Google’s AI Overviews alone may have cut publisher search referrals by 20 to 60 percent per OpenAI’s own expert), means the web’s content quality degrades over time as the economic incentive to produce it disappears. LLMs trained on an impoverished future web will be worse than the ones trained on the rich past web, and every enterprise application built on top of them inherits that degradation.
The litigation outcome is genuinely uncertain, but the strategic exposure is not. Boards and general counsels at companies deploying AI tools built on unlicensed web scraping should treat these documents as a materiality signal, not a spectator sport. Fair use as a training data defense has now been publicly undermined by the defendants’ own written record. If the Times wins, or if settlement terms create a licensing precedent, the cost structure of every major frontier model shifts, and the vendors currently offering flat-rate enterprise AI subscriptions are repricing. The budget you’re defending today was built on assumptions that may not survive the next 18 months of this case.
Concept deep-dive: The doom loop
The “doom loop” describes a specific feedback collapse: AI answers replace the need to visit publisher websites, which destroys the ad and subscription revenue that funds journalism, which reduces the volume and quality of new content being published, which degrades the training data available for the next generation of AI models. Think of it as a factory that burns its own inventory to power the assembly line. Microsoft’s internal documents named this dynamic explicitly, making it one of the rare cases where a technology company’s own strategy memo doubles as its prosecution exhibit.
Based on reporting from OpenAI and Microsoft knew they were starting a ‘doom loop’ for the web, originally published 2026-09-18 17:07:00.
