Unsealed court documents in the New York Times v. OpenAI lawsuit reveal admissions by Microsoft and OpenAI executives that large language models were trained on stolen content, creating a 'doom loop' that destroys web traffic and business models of the sources they trained on. Internal Microsoft documents acknowledge LLMs represent an 'astonishing theft of unprecedented proportions' and the 'largest theft of labor in human history,' with clicks to news sites declining over 90 percent after content was used to train AI systems.