Microsoft: Less Than 1% of Copilot Logs Used NYT

Microsoft has a number for the judge in its ongoing copyright fight with The New York Times, and it's a small one. In court filings submitted as part of discovery, the company says fewer than 1 percent of 8.2 million Copilot chat logs turned up even 16 matching words from news content the chatbot was trained on. That's the figure Microsoft is leaning on to argue that Copilot doesn't function as a substitute for reading the original articles, books, or investigative reports it was trained on.

The logs themselves weren't random. Microsoft says it handed over conversations specifically flagged for keywords tied to plaintiff websites, meaning the sample was already stacked toward finding overlap. Even so, the company says the results came back thin: 59,545 logs out of millions showed any meaningful word overlap with news content, and separate analyses tied to the Center for Investigative Reporting and a group of book authors turned up similarly small numbers.

Microsoft is now using that scarcity as evidence in its defense against The New York Times, CIR, and the Authors Guild, all of which have accused the company of building its AI on copyrighted work without permission or payment.