The legal foundation for AI training on copyrighted data is cracking under the weight of its own creators’ words. A new 92-page brief filed by The New York Times, Daily News group, and Ziff Davis in US District Court reveals internal communications that directly contradict the fair use defense mounted by OpenAI and Microsoft. The plaintiffs are seeking billions in damages, arguing that these tech giants didn’t just borrow content—they built their business models on what their own staff described as systematic appropriation.
The most damaging evidence comes from within Microsoft itself. Brent Hecht, the company’s director of applied science, labeled the practice “an astonishing theft of unprecedented proportions” and potentially the “largest theft of labor in human history.” He went further, suggesting a successful fair use defense would make a mockery of the concept entirely. While Microsoft dismissed this as one employee’s personal view, the scale of the admission is hard to ignore. Nick Turley, OpenAI’s head of ChatGPT, admitted internally that their products are “largely substitutive” and pose an “existential threat” to publishers. He noted that users simply won’t click through to original sources, no matter how prominently links are displayed.
Data backs up these candid assessments. Microsoft’s own metrics show Copilot click-through rates plummeted by 87 to 93 percent for the New York Times and up to 94 percent for Ziff Davis properties compared to traditional Bing searches. Internally, Microsoft engineers warned of a “doom loop” where AI models degrade the very web content they rely on, threatening their own supply chain. Satya Nadella testified under oath that chatbot conversations have replaced visits to original sources, though Microsoft later clarified this was about broad information consumption trends rather than a specific copyright concession.
Beyond the economic impact, the brief alleges deliberate technical circumvention. OpenAI co-founder Greg Brockman reportedly responded “ah nice” to an employee’s hack for bypassing the New York Times paywall. Testimony indicates OpenAI made no effort to review terms of service or detect paywalled content during crawling. With the US Copyright Office previously noting that fair use struggles to apply at this scale, these internal admissions strip away the veneer of accidental infringement. If courts accept that these companies knew they were replacing the source material they scraped, the precedent could reshape the entire generative AI industry.
