Unsealed court filings reveal Microsoft called AI data scraping 'theft'

Unsealed filings in The New York Times’ three-year-old copyright lawsuit against OpenAI and Microsoft show a Microsoft executive privately called the companies’ AI training practices “the largest theft of labor in human history,” according to TechCrunch.

The quote comes from a January 2023 internal memo by Brent Hecht, Microsoft’s director of applied science. A separate presentation Hecht wrote in January 2024 warned that Microsoft’s Copilot “answer engine” was pulling traffic away from the news sites that supply its training data, calling the pattern a “doom loop” that would “hurt the performance of our models and the entire web at the same time.” Microsoft’s own data, cited in the filing, showed Copilot cut click-through rates to The New York Times’ website by as much as 93% compared with traditional Bing search results.

Executives at both companies used similarly stark language about the threat their products pose to publishers. OpenAI’s head of ChatGPT, Nick Turley, wrote internally that news publishers face an “existential threat” from chatbots that are “largely substitutive” for the original journalism and will “get more and more substitutive as they get better.” OpenAI President Greg Brockman called the models “excellent at news.” Microsoft CEO Satya Nadella testified in a deposition that using a chatbot “has substituted” for visiting the original source, and said he would have invoked Microsoft’s right to require OpenAI to retrain its models had he known OpenAI trained on paywalled content.

The filings also detail the scale of the copying. OpenAI’s mid-training datasets contain more than 91,692 copies of articles from the Times, the Daily News, and the Center for Investigative Reporting, the news organizations involved in the case. A dataset built from Common Crawl, a free archive of web pages, included more than 2 million documents from nytimes.com alone. Separately, the filings say OpenAI and Microsoft shared data through internal initiatives called Project Taxi and Project Mango; the Mango dataset alone contained at least 160,903 unique works from the news publishers.

The filings also allege OpenAI employees worked out how to get past the Times’ paywall without being detected. When researcher Nick Ryder told Brockman about a “hack to get around nytimes paywall,” Brockman replied, “ah nice.” Engineers also allegedly stripped copyright notices from training data so the resulting chatbot outputs would not display them.

Much of this new detail comes from the Times’ own legal brief rather than the underlying exhibits, which remain sealed, so the quotes lack their original context. OpenAI and Microsoft did not respond to TechCrunch’s requests for comment. Courts have generally favored AI companies on whether training counts as fair use, meaning use of copyrighted work without permission, and the Trump administration filed a brief this month defending OpenAI’s position. Watch for whether the Times can use these internal admissions to undercut that fair-use defense, and whether the remaining sealed exhibits get unredacted next.