Unsealed NYT filings: Microsoft exec called AI training "theft"

Newly unredacted material in the three-year-old copyright lawsuit The New York Times brought against OpenAI and Microsoft reveals internal admissions that the companies privately viewed their own AI training practices as damaging to publishers. Per the lawsuit, a senior Microsoft executive described the companies’ AI training practices as “theft,” and OpenAI’s own leadership said its AI models posed an “existential threat” to the publishers and journalists whose work trained them. The unsealed material also alleges the companies obtained and used that content by bypassing paywalls undetected, building training datasets through mass scraping, and deliberately stripping copyright notices from the data. Much of the new material comes from the Times’ own brief rather than the underlying exhibits, which remain sealed, and the quotes are presented without their original context.

The filing is the latest escalation in the case, in which the Times alleges the companies violated copyright law by training generative AI models on its content. Courts have so far leaned toward AI companies’ “fair use” defense — the doctrine permitting copyrighted work to be used without permission in cases like parody or news reporting — and the Trump administration filed a brief earlier this month backing OpenAI’s position. But several new admissions cut against fair use’s core requirement that the use not substitute for the original work’s market: Microsoft’s own data reportedly shows its Copilot “answer engine” cut click-through rates to the Times’ domain by as much as 93% versus traditional Bing search, a decline an internal Microsoft presentation called a “doom loop.” Nadella testified that anything paywalled should be licensed by whoever wants to use it for training, and that he would have required OpenAI to retrain its models had he known it scraped paywalled content. OpenAI’s head of ChatGPT, Nick Turley, wrote internally that the chatbot poses an existential threat to publishers because it is “largely substitutive” of the original source.

The documents describe the scale of the alleged copying: OpenAI’s mid-training datasets reportedly contain more than 91,692 copies of works from the Times, the Daily News, and the Center for Investigative Reporting, and a Common Crawl-derived dataset allegedly included more than 2 million documents from nytimes.com alone. A separate dataset from a Microsoft-OpenAI data-sharing initiative allegedly contained at least 160,903 unique works from news publishers. In a January 2023 memo cited in the filing, Microsoft’s Brent Hecht reportedly called the practice “an astonishing theft of unprecedented proportions” and “the largest theft of labor in human history.”

“The evidence revealed here for the first time shows that OpenAI and Microsoft knew that what they were doing was wrong,” Steven Lieberman, counsel for the New York Daily News, said in a statement shared with TechCrunch. OpenAI and Microsoft did not return requests for comment.

Why it matters

If courts weigh these internal admissions alongside the “fair use” defense, the case could reshape how much legal cover AI companies have to train on copyrighted journalism without paying for it — a question with direct stakes for every publisher whose work already sits inside a foundation model’s training data.