OpenAI and Microsoft knew they were starting a 'doom loop' for the web

Recently unsealed court documents from the New York Times’ suit reveal that OpenAI and Microsoft internally recognized their large-language-model strategy was actively cannibalizing the web’s economics. Executives and researchers warned the companies were creating a “doom loop” in which AI products deliver answers that replace visits to news sites, undermining the referral traffic and ad revenues that fund journalism. Internal characterizations range from calling the scraping and reuse of publisher content the “largest theft of labor in human history” to conceding that the companies’ defenses would make a mockery of fair use. Microsoft and OpenAI publicly distanced themselves from some individual statements, but the filings include admissions from senior leaders acknowledging substitution of publisher content and the existential threat to creators.

The documents detail how teams ignored paywalls and terms of service while assembling training data, noted model memorization that can reproduce copyrighted text verbatim, and catalog several instances of ChatGPT outputting long passages from the Times and other outlets. Employees warned LLMs “substitute for the labor” that produces culture and news, and internal memos admitted the product competes with and can “destroy its supply chain.” Parallel business incentives are visible too, with founders and executives focused on huge commercial upside as OpenAI pursues a blockbuster valuation.