Microsoft Executive Called AI Scraping ‘Largest Theft of Labor,’ Filing Says
Unsealed Times filings allege OpenAI and Microsoft bypassed paywalls, stripped copyright notices and cut publisher traffic by up to 93% in AI searches.
Summary
Unredacted material in The New York Times’ three-year copyright lawsuit says Microsoft director of Applied Science Brent Hecht called AI scraping “the largest theft of labor in human history” in January 2023. In January 2024, he warned Microsoft’s Copilot answer engine cut Times click-through rates by up to 93% versus traditional Bing, creating a “doom loop” that could weaken model quality and the web. OpenAI’s head of ChatGPT, Nick Turley, called chatbots increasingly substitutive and an existential threat to publishers; President Greg Brockman said models excelled at news. Microsoft CEO Satya Nadella testified in 2026 that paywalled material should be licensed and said he would invoke Microsoft’s right to make OpenAI retrain if told it had trained on paywalled data. Microsoft also identified a real employment risk for workers who created training material.
The filing alleges OpenAI’s mid-training sets held more than 91,692 copies from the Times, Daily News and Center for Investigative Reporting; one Common Crawl-derived set held over 2 million nytimes.com documents; and Project Mango contained at least 160,903 unique publisher works. OpenAI allegedly scraped the Bing Index, used Nick Ryder’s paywall “hack,” welcomed by Brockman, overrelied on news in WebText and WebText2, drew millions of Common Crawl articles and removed copyright notices. OpenAI gave Microsoft the full GPT-3 training set; Microsoft supplied OpenAI through Project Taxi and Project Mango. Judges have largely favored AI companies’ fair-use arguments, and the Trump administration backed OpenAI’s unlicensed training earlier in September 2026, but substitution and market harm could weaken that defense. Much of the disclosure comes from the Times’ brief, with underlying exhibits sealed and quotations lacking original context. OpenAI and Microsoft did not comment.
Positives
- Satya Nadella testified that paywalled material should be licensed and said Microsoft could require OpenAI to retrain models built with it.
- The unredacted filing quantifies previously undisclosed copying across mid-training datasets, Common Crawl and Project Mango.
- Brent Hecht’s January 2024 analysis recognized that starving publishers of traffic could also damage Microsoft’s models and the wider web.
- Judges and the Trump administration have supported OpenAI’s argument that unlicensed AI training can qualify as fair use.
Risks & concerns
- Microsoft data showed Copilot reduced click-through rates to The New York Times by as much as 93% compared with traditional Bing.
- Nick Turley described chatbots as increasingly substitutive and an existential threat to publishers whose journalism trained them.
- OpenAI datasets allegedly contained over 91,692 publisher copies, more than 2 million Times documents and at least 160,903 Project Mango works.
- OpenAI allegedly bypassed paywalls, scraped the Bing Index and Common Crawl, and removed copyright notices before training.
- Microsoft identified a real risk that generative AI could displace the workers responsible for producing its training material.
- Underlying exhibits remain sealed, leaving key quotations without their original context as the three-year lawsuit continues.