Internal Documents Reveal: OpenAI and Microsoft Knew They Were Triggering a "Vicious Cycle" for the Internet

Recently unsealed court documents from The New York Times' lawsuit against OpenAI and Microsoft shed an uncomfortable light on how the two companies privately viewed the consequences of their own data-collection practices used to train artificial intelligence models. According to The Verge, the companies' internal records contained explicit warnings about the risks this approach posed to the web ecosystem as a whole.

A "Vicious Cycle" Acknowledged From Within

According to materials cited by The Verge, employees at both companies used the term "doom loop" to describe the phenomenon triggered by their large-scale content-scraping practices. The underlying idea is that uncontrolled extraction of information from websites—used to power language models like the ones behind ChatGPT—gradually undermines the economic viability of publishers and creators of original content, the very people who produce the material these systems learn from.

In practice, as users grow accustomed to getting answers directly from chatbots instead of visiting source websites, traffic to publishers declines, advertising revenue erodes, and their ability to keep producing quality journalism or content is put at risk. This decline, in turn, could even affect the quality of data available for future generations of AI models, creating a cycle that feeds on itself negatively.

Harsh Accusations, From Within the Companies Themselves

The disclosed documents contain particularly blunt language. According to The Verge, one of the unsealed passages describes the scraping practices as "the greatest theft of labor in human history"—a statement all the more notable given that it comes from inside the very companies targeted by the lawsuit, not from the plaintiffs.

Another cited excerpt suggests that the way the concept of "fair use" was applied to the collected data "makes a mockery" of that legal principle—an allusion to how the companies allegedly stretched legal boundaries to justify the use of copyrighted material.

Implications for the Ongoing Case

These documents could strengthen The New York Times' position in the pending litigation, offering evidence that employees at the accused companies were aware of the problematic—perhaps even harmful—nature of the practices they were implementing. The case remains one of the most closely watched lawsuits in the field of artificial intelligence, with the potential to set important precedents regarding the use of copyrighted content in training language models.