Unsealed NYT v. Microsoft/OpenAI filings: a Microsoft scientist predicted the public would call AI training theft

On September 17, 2026, newly unredacted filings became public in The New York Times Company v. Microsoft Corporation, et al., the copyright suit The Times filed against Microsoft and OpenAI on December 27, 2023 in the U.S. District Court for the Southern District of New York. Among the documents is internal writing by Brent Hecht, Microsoft's Director of Applied Science. In a January 2023 memo, Hecht told colleagues what he thought was coming: "millions of people around the world will soon consider large models 'hoovering up' all their work to be an astonishing theft of unprecedented proportions." The filings also quote him raising the possibility that the practice amounted to the largest theft of labor in human history, and describing the resulting AI models as "a product that destroys its supply chain." A year later, in a January 2024 internal presentation, he laid out a "doom loop": Copilot pulls readers away from news sites, those sites publish less, and the next round of models has less to learn from.

The January 2023 memo reads as market research: this is how the people whose work we are using are going to describe what we are doing, and you should plan for it. Microsoft has said the comments reflect one employee's individual perspective and do not represent the company's views. What makes the memo worth reading is that the forecast was right, and that it was filed inside the company years before a courtroom gave anyone a reason to write it down.

He was not the only one. Satya Nadella, Microsoft's CEO, testified that paywalled content "should be licensed" for training, and that had he known about the paywall scraping at the time, he would have used Microsoft's contractual right to require OpenAI to retrain. He also acknowledged that chatbots "substituted" for visiting the original websites, which is the plain-language version of the Times's core legal claim. Nick Turley, who leads ChatGPT at OpenAI, wrote that publishers face an "existential threat" from AI products that are "largely substitutive." Greg Brockman, OpenAI's president, called the models "excellent at news" and replied "ah nice" when a researcher described a plan to get around paywalls.

The filings also put numbers on the practice. Per TechCrunch, OpenAI's training datasets contained more than 91,692 copies of works from The Times, The Daily News, and the Center for Investigative Reporting. More than 2 million documents from nytimes.com turned up in the Common Crawl dataset used for training. An internal dataset called "Project Mango" held more than 160,903 unique news publisher works. Internal Microsoft figures put Copilot's answer engine at up to 93% fewer click-throughs to the Times's domain than traditional Bing search returned for the same questions. That figure compares two ways of answering. It does not track what happened to the Times's traffic after Copilot launched. The filings allege that employees deliberately stripped copyright notices from the training data before it went in.

What makes this filing different from the usual AI copyright story is where the warning came from. No plaintiff's lawyer wrote it and no outside critic wrote it. It came from an applied-science director inside the company now being sued. His January 2023 memo told colleagues that the people whose work was being used would come to call it theft. His January 2024 presentation told them that products built this way would starve the sources they run on. The work continued regardless.

Strip the legal argument away and look at what actually broke. A system trained on someone else's reporting produced answers that substituted for that reporting, without carrying any pointer back to where it came from, without compensation, and, per the filings, with the original attribution deliberately removed. That is not a side effect of how large language models work. It is a design choice about whether an output is required to carry a traceable link back to a checkable source. Nadella's own testimony makes the point without meaning to: he says he would have ordered a retrain if he had known. The fix he is describing after the fact is exactly the kind of requirement that belongs at the start of a system's design, not as a correction issued once a lawsuit forces the question.

That is the design bet behind ForIntel, Foragentis's knowledge-products line. Every report we produce is built so that a claim the reader cannot trace back to a named, checkable source does not make it into the finished report. Provenance is a structural requirement of how the report gets assembled, not a footnote added after the fact or a fix bolted on once someone objects.

How the court rules on damages, fair use, and the DMCA claims in this case is a separate question from the one worth asking now: was the extraction-without-attribution the only way to build a useful AI system? The people building it wrote down their own doubts about that years ago. If you are evaluating an AI vendor and want to know whether their outputs trace back to real, checkable sources, that is exactly the question ForIntel was built to answer: https://foragentis.com