Microsoft exec called AI scraping the “largest theft of labor in human history”

Fair-use fights often sound abstract. These documents don’t.

Hecht’s memos call scraping “astonishing theft” and say mass news training makes “a complete mockery” of fair use. OpenAI’s Nick Turley calls publishers existentially threatened; Microsoft writes of a “doom loop” that hurts models and the web together. Measured click-through drops for some plaintiffs hit 83–93%.

Microsoft now says Hecht spoke for himself. News plaintiffs say the cat’s out of the bag—and want summary judgment on extensively overlapping verbatim outputs.

We’re reading the unsealed motion as the industry’s private voice colliding with its public fair-use story.

Nadella under oath: paywalled material should be licensed; chatbots steal clicks by answering on the AI surface. Brockman on a paywall “hack”: “Ah, nice.” Turley: no good reason to click when the bot already answered; products get more substitutive as they improve. Plaintiffs stress substitution plus verbatim excerpts from summaries, bias ratings, and homepage scrapes.

Microsoft’s spokesperson frames Nadella as observational and Hecht as one employee. Counsel for Daily News papers calls the secrecy the tell. Separate allegations—Bing crawl data resold into training, a third-party NYT dataset used despite commercial limits, filters Hecht feared looked like an “accidental cover up”—will fight at trial if summary judgment doesn’t end the core claim.

Useful AI needs a living news supply chain. A doom loop that starves publishers starves models too. I’m watching the court’s take on substitution evidence—and whether licensing finally becomes the default rather than the press-release exception.

Ars Technica, September 17, 2026. Coverage of unsealed summary-judgment materials in the NYT-led case against Microsoft and OpenAI.