Unsealed filings show Microsoft executive called AI training scraping 'theft of labor'

TechCrunch reports that newly unredacted material in The New York Times' copyright lawsuit against OpenAI and Microsoft, now three years old, shows internal views that bear on the companies' fair use defense. A senior Microsoft figure privately called AI training practices theft, and OpenAI leadership described its models as an existential threat to publishers. The article cautions that most of the detail comes from the Times' own brief rather than the underlying exhibits, which stay sealed, so the quotes appear without their original context.

According to the filing, Microsoft data showed its Copilot answer engine cut click-through rates to the Times' domain by as much as 93 percent compared with traditional Bing. A January 2024 presentation by Brent Hecht described a doom loop, and a January 2023 memo called the scraping an astonishing theft, adding that it was the biggest labor theft ever. Satya Nadella testified that paywalled content should be licensed, and said he would have required OpenAI to retrain had he known it used such material.

The filing also alleges how content was gathered. OpenAI's mid-training data held more than 91,692 copies of works from the Times, Daily News and the Center for Investigative Reporting, and one dataset drawn from Common Crawl had over 2 million nytimes.com documents. The filing says OpenAI handed Microsoft its whole GPT-3 training dataset, and that Microsoft supplied OpenAI with training data under two programs, Project Taxi and Project Mango. The Mango data was allegedly assembled into a dataset with at least 160,903 unique news works. An OpenAI researcher reportedly told Greg Brockman of a paywall workaround, and he replied approvingly. The article adds that judges have mostly favored fair use arguments so far, and that Microsoft and OpenAI did not comment.