A recent court filing has revealed that employees at OpenAI and Microsoft are aware that their artificial intelligence products compete directly with news organizations whose articles were used to train these models. According to the document, which was made public by The New York Times and other publishers, OpenAI’s head of ChatGPT, Nick Turley, commented that their products are fundamentally “substitutive”—this seems to suggest they could replace, or at least diminish, the role of traditional news outlets as the technology evolves.
Brent Hecht, a Director of Applied Science at Microsoft, expressed concerns that many people may view the large models as “hoovering up” all their work, likening it to an unprecedented theft. This perspective appears to contradict the Department of Justice’s argument that AI training constitutes fair use, as it suggests models learn from text patterns rather than replicating the content itself—something that internal communications appear to undermine.
In the document, Turley also pointed out that news publishers are facing an “existential threat” due to the rise of these AI products. Additionally, one document referenced described the situation as potentially the “largest theft of labor in human history.” Hecht further noted that if the companies were to win on the fair use front, it might lead to a situation that completely undermines the very concept of fair use itself.
Microsoft and OpenAI did not respond immediately to requests for comments from the DCNF. A tech policy expert observed that the statements made within these communications contradict the companies’ legal defenses. Evan Swarztrauber, who previously advised FCC Chairs, mentioned that a significant aspect of fair use is whether the copied product competes with the original, and these internal discussions highlight a clear acknowledgment of substitution.
Microsoft’s documents also reflected similar apprehensions regarding the impact of AI on publishers. One internal document indicated that their AI content strategy was leading to a “doom loop” that could damage both their models and the broader web ecosystem. In essence, it seemed to present a scenario where the product they’ve developed threatens the economic base of those who supply essential content.
Another quote within the brief described large language models as “a product that destroys its supply chain.” Microsoft CEO Satya Nadella noted that interacting with chatbots has effectively replaced the need to visit original sources and suggested that anything behind a paywall should require licensing for AI training. He also mentioned he would have insisted that OpenAI retrain its models if he had been aware they were using paywalled content.
Greg Brockman, OpenAI’s co-founder and President, indicated that their models excel at predicting the text of news articles. When an employee, Nick Ryder, brought up a way to bypass The New York Times’ paywall, Brockman’s response was simply, “ah nice.” The corporate representative for OpenAI testified that he was unaware of any efforts to identify or remove paywalled material from their training data.
A Microsoft spokesman stated that Hecht’s comments represented the views of just one employee and did not reflect Microsoft’s position on the matter. OpenAI has claimed that using publicly available materials for training purposes falls under fair use, as articulated in statements made earlier this year.
In July 2025, former President Donald Trump commented on the necessity of having access to a wide range of materials without imposing fees for successful AI development during the announcement of the AI Action Plan.
Swarztrauber argued that this case might encourage the creation of a market for licensing the work of publishers, something he believes has been suppressed by the AI industry’s position that they owe nothing to content creators. “There’s no reason that the most sophisticated companies in human history cannot figure out a way to pay people for their work,” he stated.
