Newly unsealed court documents in the New York Times’ lawsuit against OpenAI and Microsoft reveal a stark contradiction between public defenses and internal reality. The filings show that top executives and researchers at both companies privately acknowledged their generative AI models were built on what one Microsoft director called the “largest theft of labor in human history.” Brent Hecht, Microsoft’s Director of Applied Science, explicitly stated that the companies’ fair use arguments made a “complete mockery” of legal precedents. While Microsoft now claims Hecht’s views were merely “academic” and not representative of corporate strategy, the sheer volume of corroborating internal memos suggests this was a known operational risk, not a rogue opinion.
The most damaging admission concerns the economic sustainability of the web itself. An internal Microsoft document describes the current AI strategy as a “doom loop” that threatens the very content supply chain it relies on. By replacing search referrals with direct answers, chatbots remove the incentive for users to visit source websites. OpenAI’s Nick Turley reportedly noted that once a user gets an answer from ChatGPT, there is “no good reason to click” through to the original article. This behavior is already having measurable effects, with internal experts estimating that search referrals for major publishers could drop by as much as 60 percent. Satya Nadella even admitted that chatbots have effectively replaced traditional search, stripping publishers of their primary traffic engine.
Technical teams were equally candid about the models’ limitations and legal exposures. Employees acknowledged that GPT-4 was “insanely good at regurgitation,” memorizing vast amounts of copyrighted text despite efforts to prevent it. The documents cite specific instances where ChatGPT reproduced long strings of copy from the Times, Mercury News, and other outlets verbatim. Furthermore, while Nadella publicly stated that paywalled content should be licensed, an OpenAI representative admitted internally they were “unaware” of any effective mechanism to detect or remove paywalled material from training datasets. This gap between public rhetoric and private capability undermines the argument that these companies acted in good faith.
These revelations strip away the veneer of accidental infringement. The data shows a calculated decision to prioritize speed to market over licensing agreements, driven by the potential for “gazillions” in revenue as noted by cofounder Greg Brockman. If the courts accept that these companies knowingly destroyed their own supply chain while dismissing copyright concerns as a “mockery,” the precedent could reshape how intellectual property is valued in the digital age. The question is no longer whether AI hurts publishers, but whether the legal system will penalize those who built their empires on that destruction.
