Unsealed Court Filings Expose Microsoft and OpenAI Fears Over AI’s Impact on News

Newly unsealed filings in The New York Times’ copyright case portray Microsoft and OpenAI executives privately acknowledging that AI systems were built with large volumes of journalistic work and could displace the publishers that produced it. A Microsoft executive described the practice as “an astonishing theft of unprecedented proportions” and “the largest theft of labor in human history,” according to the filing and multiple accounts of its contents.

The documents also describe internal recognition that products such as ChatGPT and Microsoft Copilot competed directly with news websites. OpenAI executive Nick Turley wrote that the company’s products were “largely substitutive,” while Microsoft CEO Satya Nadella testified that conversations with chatbots had replaced visits to publisher sites, Bloomberg Law reported. The statements sit alongside Microsoft and OpenAI’s public legal position that model training qualifies as lawful, transformative fair use.

The filings allege that Microsoft and OpenAI used restricted Times material in datasets, bypassed paywalls, and stripped copyright notices. The material reportedly includes more than 91,000 copies of works from The Times and related publishers in OpenAI mid-training datasets, according to a detailed account cited by Msukhareva.substack.

Those allegations matter because the case concerns both model training and the products built on that training. Publishers argue that copying articles without permission supplied the raw material for systems that can summarize or reproduce reporting, while chatbot answers reduce the clicks that fund the original work. Ground cited Microsoft data showing Copilot cut New York Times referrals by 93%, although the broader filings and publisher arguments frame traffic substitution as part of a wider economic threat.

The documents describe a feared cycle in which AI tools consume journalism, answer questions without sending readers to publishers, and weaken the revenue base needed to produce new reporting. OpenAI and Microsoft personnel reportedly recognized that the systems could threaten news organizations and journalists even as the companies defended their data practices in court.

That concern reaches beyond copyright ownership. News publishers depend on subscriptions, advertising, licensing, and referral traffic, while generative AI systems can deliver an answer inside a closed interface. If readers stop visiting the original article, publishers lose both audience data and commercial opportunities. Ars Technica described the internal concern as a “doom loop,” and Bloomberg Law reported that the admissions extended to books as well as newspaper articles.

The statements give publishers evidence for an argument that the AI companies understood the commercial effect of their systems and the value of the material used to train them. They also help plaintiffs connect data collection, product design, and traffic loss into one claim about market harm.

The admissions don't decide whether the companies infringed copyright. Courts still must assess questions including fair use, the legal significance of intermediate copying, the nature of model outputs, and whether publishers can prove measurable economic injury. Microsoft and OpenAI maintain that their AI products are transformative, while the newly public communications provide the publishers with a sharper contrast between private assessments and public defenses.

The filings increase pressure on Microsoft and OpenAI to explain how training data was obtained, how copyrighted works were handled, and whether safeguards prevented verbatim reproduction. They also raise questions for other model developers, because the dispute targets a business practice shared across the generative AI industry rather than a single chatbot feature.

For publishers, the case strengthens the push for licensing agreements, attribution, crawler controls, and compensation tied to AI usage. For developers, it reinforces the need for documented data provenance and systems that distinguish licensed content from material gathered through access restrictions. The dispute has already shifted from an argument about technical training methods to a fight over who captures the economic value of professional knowledge.

The next stage will turn the newly unredacted evidence into arguments over summary judgment and, if necessary, trial. The New York Times and other publishers will use the executive statements, dataset figures, and traffic data to argue that the companies knew their systems substituted for original reporting and threatened the market for it.

Microsoft and OpenAI will contest the legal weight of those statements and the interpretation of the underlying data. The outcome will shape how courts evaluate AI training, publisher licensing, chatbot citations, and claims that generative systems replace visits to original websites. It will also influence negotiations between model developers and news organizations, which now have a clearer record of the economic risk at the center of the fight.