Internal Documents Reveal AI Giants Knew Models Would Replace News
Newly unsealed court filings show that senior executives at OpenAI and Microsoft explicitly warned their leadership that AI products would compete directly with news publishers, a fact that complicates their legal defense.
Internal communications from OpenAI and Microsoft, recently made public in a court filing, indicate that company leaders understood their artificial intelligence systems were poised to replace journalism rather than merely assist it. These documents, reviewed by reporters for GN technics/ai (en-US), suggest that the companies were aware of the economic impact on publishers while simultaneously building products that draw users away from traditional news sites.
The core of the dispute lies in whether using copyrighted news content to train AI models constitutes fair use or unauthorized theft. Publishers argue that the AI systems now serve as direct substitutes for their websites, effectively stealing the value of their work. The newly revealed emails and memos provide evidence that the technology firms recognized this substitution effect before launching their consumer-facing products.
Executives warned of existential threat
Nick Turley, the head of ChatGPT at OpenAI, reportedly described the situation as an "existential threat" to publishers. His internal notes stated that AI products were already "largely substitutive" and would become even more so as their capabilities improved. This assessment contradicts the public narrative that AI tools are complementary to human creativity, instead framing them as competitive replacements for the labor of journalists.
Brent Hecht, a Director of Applied Science at Microsoft, used even stronger language in his internal communications. He described the alleged copying of news content as an "astonishing theft of unprecedented proportions." He went so far as to call it perhaps the "largest theft of labor in human history," highlighting the scale of the data extraction involved in training these large language models.
Data shows declining publisher traffic
Microsoft’s own internal data, cited in the filings, appears to support the publishers' claims of harm. The documents describe a "doom loop" where the AI strategy hurt both the performance of the models and the broader web ecosystem. Traffic analysis showed that click-through rates from Bing Chat to news websites were significantly lower than traditional searches, with drops ranging from 83% to 93% for several major publications.
Satya Nadella, Microsoft’s CEO, testified that users increasingly rely on chatbots for information, eliminating the need to visit source websites. He acknowledged that paywalled material should be licensed for AI training and agreed that accessing data in violation of terms of service is improper. However, he also stated that if he had known OpenAI scraped paywalled content, he would have required the company to retrain its models, suggesting a potential disconnect between policy and practice.
Legal defense rests on fair use
Despite the internal warnings, OpenAI and Microsoft maintain that their actions are protected under the fair-use doctrine. They argue that training AI systems on copyrighted material is a transformative use that benefits society. OpenAI has specifically accused the New York Times of seeking an "undeserved payday" at the expense of technological progress, framing the lawsuit as an attempt to stifle innovation.
The publishers counter that the AI companies circumvented paywalls and jointly scraped the internet for training material without permission. The case hinges on whether these actions constitute infringement or a legitimate part of technological development. The court has not yet ruled on the merits of these arguments, but the unsealed documents have added significant weight to the publishers' narrative of deliberate competition.
