Microsoft Insiders Call AI Training Theft of Labor
Court-ordered disclosures in the New York Times case show Microsoft and OpenAI staff privately calling AI training an unprecedented theft.
In short
Documents produced in the New York Times lawsuit show senior Microsoft and OpenAI staff describing model training on copyrighted text as theft, even as both companies publicly leaned on fair use.
At a glance
- Microsoft executive Brent Hecht privately called the practice the largest theft of labor in history.
- OpenAI ChatGPT lead Nick Turley wrote in June 2023 that the product is an existential threat to publishers.
- The New York Times sued in late 2023; the consolidated case runs as No. 1:25-md-03143-SHS-OTW.
- An internal Microsoft paper describes a vicious cycle in which the AI strategy erodes its own training material.
- Hecht argued that stretching fair use over this practice would make the term ridiculous.
Discovery in the New York Times case against OpenAI and Microsoft has pushed internal messages into the court record, and they read nothing like the companies public filings. Executives at both firms describe training on other peoples text as a taking, not a gray area. German outlets heise online and t3n reported the same passages.
The quote at the center of it
Brent Hecht, who leads applied science at Microsoft, is the bluntest voice in the record. He called the unlicensed use of protected text a theft of unprecedented scope, and floated that it may be the largest theft of labor in history. Bending fair use to cover it, he added internally, would make the doctrine ridiculous.
Nick Turley, who runs ChatGPT at OpenAI, made a narrower point in June 2023 with similar force. In his reading the assistant mostly substitutes for press coverage rather than sending readers toward it, which he framed as an existential threat to publishers.
A loop that eats its own inputs
A separate document from Microsofts legal side sets out the business logic. Chatbot answers cut clicks to original reporting, and thinner newsroom revenue eventually thins the supply of the material models are trained on. The memo treats that as a risk to model quality, not only to the press.
Why it matters in court
The Times filed in late 2023, alleging that millions of its articles went into training without permission, and the consolidated matter is docketed as No. 1:25-md-03143-SHS-OTW. US fair use analysis weighs, among other things, whether a use damages the market for the original work. Statements in which defendants own staff describe substitution and existential harm hand the plaintiff that argument in the defendants words.
What the record does not settle
An employee email is evidence, not a corporate legal position, and it is not a ruling. The reporting reviewed here contains no fresh response from Microsoft or OpenAI to the quoted passages. A third account, from derStandard.at, was readable only down to its paywall, so its details could not be cross-checked. The case remains undecided.
FAQ
What do the internal Microsoft documents say about AI training?
They show executives describing the use of copyrighted text as theft of unprecedented scope, and warning that a fair use defense would strain the doctrine past credibility.
Who is Nick Turley and what did he write in June 2023?
Turley leads ChatGPT at OpenAI. He wrote internally that the product largely replaces press coverage instead of driving readers to it, calling that an existential threat to publishers.
Does this decide the New York Times lawsuit?
No. The filing dates to late 2023 and the consolidated case, No. 1:25-md-03143-SHS-OTW, has not been decided; these documents are evidence a court still has to weigh.
