OpenAI Microsoft Copyright Lawsuit Exposes News Scraping

OpenAI and Microsoft face increased legal and commercial risk after plaintiffs unsealed summary-judgment filings on Sept. 17, 2026, revealing internal messages that acknowledge large-scale news scraping and product substitution. Plaintiffs argue these admissions weaken the companies’ fair-use defense in the ongoing copyright lawsuit.

Unsealed Filings Reveal Internal Admissions

The disclosures come from summary-judgment papers filed by news plaintiffs, including The New York Times, in the consolidated multidistrict litigation In re OpenAI Inc. Copyright Infringement Litigation (S.D.N.Y., No. 1:25-md-03143). The Times initiated the case in 2023, which was later consolidated with other suits. The newly unsealed material consists mainly of plaintiffs’ briefs and discovery excerpts such as internal emails, memos, and deposition testimony, not judicial findings.

A January 2023 memo from Microsoft director Brent Hecht described large-scale news scraping as “an astonishing theft of unprecedented proportions” and warned it could trigger a cycle damaging both model performance and the broader web. Microsoft said Hecht’s remarks reflected one employee’s view and were not a legal analysis of company policy.

The filings quote OpenAI’s head of ChatGPT describing the pressure on publishers as severe and warning that AI tools already substitute for journalistic work, a trend expected to increase as AI improves. OpenAI co-founder Greg Brockman acknowledged the models’ aptitude for news tasks and responded approvingly when told of a workaround to bypass a paywall.

Deposition excerpts from Microsoft Chief Executive Satya Nadella show him recognizing that conversational AI can replace visits to original news sites. He testified he would have invoked Microsoft’s contractual right to require retraining if informed that models were trained on paywalled content. Plaintiffs contend these internal acknowledgments undermine the companies’ broader claim that training on published text is non-substitutive and thus fair use.

Dataset Scale and Legal Stakes

Plaintiffs allege OpenAI and Microsoft copied millions of copyrighted articles to train GPT-3, ChatGPT, and Microsoft Copilot, with OpenAI scraping more than 10 million articles overall, nearly one-third from The New York Times.

The filings identify mid-training datasets containing over 91,692 copies of works from The New York Times, New York Daily News, and the Center for Investigative Reporting. A Common Crawl-derived subset reportedly included more than 2 million nytimes.com documents. A joint Microsoft–OpenAI project called Project Mango assembled a corpus with at least 160,903 unique news works.

Plaintiffs claim OpenAI provided the full GPT-3 training dataset to Microsoft for internal evaluation, and Microsoft returned data through initiatives named Project Taxi and Project Mango. The filings say OpenAI relied heavily on datasets labeled WebText and WebText2, drew millions of documents from web crawls and search indexes, and removed publisher copyright notices before incorporating material into training sets.

The plaintiffs cite Microsoft data showing Copilot reduced click-through to nytimes.com by roughly 93% compared with traditional Bing search. They seek damages tied to each allegedly infringed article, with some reports estimating potential exposure in the billions of dollars.

Following a Supreme Court ruling involving Cox Communications, plaintiffs reframed contributory-infringement claims to emphasize Microsoft’s role in designing systems that trained on Times content.

Microsoft and OpenAI maintain that using publicly available text to build new, transformative models qualifies as fair use. They cite recent federal rulings, including Kadrey v. Meta and Bartz v. Anthropic, that found training on copyrighted works can be transformative. The U.S. Department of Justice filed a statement of interest urging courts not to restrict training practices that would hinder innovation. Microsoft characterized Nadella’s testimony as commentary on how people find information, not a legal conclusion about copyright. Microsoft also said Hecht’s statements represent one employee’s perspective, not company policy.

Magistrate Judge Ona Wang disclosed in a Sept. 14 filing that she had divested Microsoft stock and would remain on the MDL, noting reassignment would not serve the public interest.

Legal Outlook and Next Steps

District Judge Sidney Stein is expected to rule in 2027 on which claims in the Times-led case will proceed to trial. The MDL’s outcome could shape when and how publishers’ material may be used in generative AI training and influence licensing and settlement dynamics between news organizations and AI companies.