Unsealed Court Documents Show OpenAI and Microsoft Warned of a Web 'Doom Loop'

Internal documents from OpenAI and Microsoft, unsealed in the New York Times' copyright lawsuit against both companies, show that their own employees and advisors anticipated the damage large-scale AI scraping would do to the web — and used unusually blunt language to describe it. According to records made public in the case, the companies warned they were setting off a "doom loop" that would degrade the open web, and characterized their scraping of human-created content to train models as "the largest theft of labor in human history." The filings were reported by The Verge.

The 'doom loop' framing describes a self-reinforcing cycle: AI systems scrape human-written content, those creators lose traffic and revenue, less original material gets produced, and future models end up training on their own synthetic output. Each turn of the loop degrades the quality of what is available to train on. The documents suggest the companies understood that dynamic in advance rather than discovering it after the fact.

The New York Times sued OpenAI and Microsoft in December 2023, alleging that the companies used millions of the paper's articles to train their models without permission or payment. The case has become one of the highest-stakes tests of how U.S. copyright law applies to generative AI, and it has forced both defendants to hand over internal communications. Those communications are now surfacing publicly, and they read less like a legal defense than an acknowledgment of the problem.

The language in the records is notable for how directly it frames the economic stakes. Calling the practice "the largest theft of labor in human history" is not the sort of phrase a company would put in a press release — it is the kind of internal candor that plaintiffs in copyright cases look for. The Times has argued that OpenAI and Microsoft built commercial products on content they did not license, and documents in which the defendants themselves describe the extraction as theft give that argument more weight.

The unsealed material lands as courts and regulators are already weighing how training data should be sourced and compensated. Several publishers have struck licensing deals with AI developers, while others have sued, and the outcomes will shape whether scraping remains free or becomes a paid input. Internal warnings about a doom loop could also factor into arguments about willful infringement, which carries heavier penalties than accidental copying.

What happens next depends heavily on how the court treats these documents. If the judge allows them as evidence of intent, they could influence both the Times case and the broader set of lawsuits against AI developers. The open question is whether the companies' own predictions about the web's decline will be read as foresight worth penalizing, or as an industry-wide problem that copyright law was never built to address.