Internal communications from OpenAI and Microsoft described their AI training practices as the "largest theft of labor in human history" and warned the companies were launching a "doom loop" that would damage the web, according to recently unsealed court documents in the New York Times' lawsuit against the two tech giants. The 92-page filing reveals that employees and executives at both companies recognized their AI models posed an existential threat to publishers, yet proceeded anyway. Many of the most striking statements come from Brent Hecht, Microsoft's Director of Applied Science, though the company has sought to characterize his views as personal opinion rather than corporate position.

The documents reveal that Microsoft admitted its AI content strategy created a "doom loop" that would simultaneously degrade the performance of its models and harm the entire web. An internal Microsoft memo noted it was "highly unusual that an end-product threatens the economic foundations of its essential suppliers," but acknowledged that's exactly what the company had done with its large language model business and its content supply chain. OpenAI employees conceded that GPT-4 "memorized a ton of data and therefore will be insanely good at regurgitation," despite recognizing that preventing memorization was crucial to minimize copyright violations. The filing includes multiple examples of ChatGPT reproducing long passages verbatim from Times articles, as well as content from the Mercury News, Denver Post, LifeHacker, and Eurogamer. An OpenAI representative admitted being "unaware" of any effort to identify or strip out paywalled content from training datasets, despite CEO Satya Nadella later stating that paywalled material should be licensed.

Microsoft's own analysis found that the company was "hoovering up" content that creators never intended to be used for AI training and for which they received no payment. OpenAI Policy Director Jack Clark warned the company was building systems that replace the work of people who shape society's culture. Internal documents labeled ChatGPT as "the modern newsstand," while OpenAI's Head of ChatGPT argued that once users receive an answer from the chatbot, there's "no good reason to click" through to the original source. OpenAI cofounder Greg Brockman appeared more focused on the "gazillions" of dollars commercial AI could generate than on addressing these concerns. Microsoft characterized Hecht's assertions as reflecting "divergent, academic, and forward-looking views" from an employee hired specifically to bring "asymmetrical, futuristic, and academic points of view," stating his comments don't represent the company's official stance.

The filing shows both companies understood they were destroying their own supply chain while irreparably harming publishers and the millions they employ. Microsoft admitted that large language models are products that destroy their own supply chains because they substitute for the very training data they require. OpenAI's media and economic specialists attributed declines in referral traffic for outlets like the Times directly to AI summaries such as Google's AI Overviews, speculating that search referrals may have dropped as much as 60 percent. Microsoft attempted to draw a distinction between Nadella's testimony about broad shifts in how people find information and the copyright questions before the court, but the unsealed documents suggest both companies knowingly proceeded despite understanding the damage they would cause. The incentive structure remains unchanged—commercial AI promises massive financial returns even as it cannibalizes the content ecosystem that makes those models possible. Whether courts ultimately side with publishers or tech companies, the fundamental tension between AI's economic model and the survival of its content sources appears destined to reshape both industries in ways neither may fully control.