Court filings reveal OpenAI and Microsoft knew they were triggering a "doom loop" for the web

Internal documents surfaced in the New York Times' lawsuit against OpenAI and Microsoft show both companies were fully aware of the damage their models inflict on the web economy — and proceeded anyway. One memo describes the training-data approach as "the largest theft of work in human history" and says the companies' legal defense makes "an absolute mockery of the concept of fair use." The sharpest language comes from Brent Hecht, Microsoft's director of applied science, who warns of "Google Zero," a state in which chatbots replace search and eliminate the need to visit the source.

Microsoft moved quickly to distance itself from Hecht's remarks. Spokesperson Alex Haurek told The Verge that the comments "reflect the personal perspective of one employee, are not legal analysis, and do not represent the company's position." In a separate filing, Jordan Usdan, Microsoft AI's director of data strategy and operations, characterized Hecht's role as that of a deliberate dissenter: he is employed to supply "asymmetric, futuristic, and academic perspectives" on data ecosystems, not to speak for Microsoft on theoretical impacts to content creators.

Whether Microsoft embraces the assessment or not, it has materialized. An internal company document warns explicitly: "Our AI content strategy has triggered a 'doom loop' that will hurt both our model performance and the web at large: it is highly unusual for an end product to threaten the economic foundations of its essential suppliers, but that is the situation we have created for our LLM business relative to its 'content supply chain.'" Satya Nadella acknowledges that chatbots have effectively replaced search and removed the need to go directly to the source.

Don't mistake the motives for altruism. Greg Brockman, an OpenAI co-founder, is documented as eyeing the "gazillions" of dollars to be made from commercial AI. And while Nadella is quoted saying "everything behind a paywall should be licensed," an OpenAI representative admitted to being "not aware" of any effort to locate or strip paywalled material from the training corpus. The gap between public posture and internal practice is stark.

Internally, OpenAI knew GPT-4 had "memorized a ton of data and therefore will be insanely good at regurgitation" of copyrighted material. Employees acknowledged that preventing memorization mattered for "minimizing copyright violations," yet the model shipped regardless. The legal filing cites concrete examples: the chatbot emits long verbatim strings from Times articles, the Mercury News, the Denver Post, Lifehacker, and Eurogamer in response to simple prompts. Microsoft, for its part, conceded that "almost no one intended" their works to be ingested this way — and kept going.