Microsoft says Copilot rarely matches NYT articles in filings
Microsoft legal filings show fewer than 1 percent of 8.2 million Copilot chat logs regurgitated at least 16 words matching New York Times articles.
Stock photo for illustration only, not from the actual event
- Microsoft shares analysis of 8.2 million Copilot chat logs in ongoing copyright lawsuit.
- Fewer than 1 percent of logs reproduced at least 16 words from publisher content.
- Company argues these figures support its fair use defense for AI model training.
Microsoft has stated in new legal filings that its Copilot AI assistant rarely reproduces full sentences or substantive chunks from news articles and books. The disclosure comes as the tech giant defends itself against copyright infringement lawsuits brought by major publishers including The New York Times and various book authors.
As part of the legal discovery process, Microsoft provided 8.2 million Copilot chat logs to an expert hired by news publishers. The company noted that these logs were specifically selected because they targeted keywords related to the plaintiffs' websites, representing the scenarios most likely to contain protected works.
Stock photo for illustration only, not from the actual event
The resulting analysis revealed that 59,545 of those conversations shared at least 16 words with news content used to ground the AI model. Additionally, an expert for the Center for Investigative Reporting found 51 instances of substantial overlap with their work, while an author-suit expert discovered only 24 responses containing 30 or more matching words across the entire dataset. Out of 212 books evaluated, only 10 showed any matches.
"hardly undermines the transformative purpose of LLM training."
Microsoft
This legal battle highlights the core tension in generative AI copyright disputes regarding fair use doctrine and transformative purpose. By demonstrating an extremely low rate of verbatim text reproduction, Microsoft aims to prove that large language models serve fundamentally different functions than original journalistic works and do not act as mere substitutes for human-authored articles.
Microsoft submitted these findings as it urges the judge to issue a summary judgment to end the case early. Representatives for The New York Times, the Center for Investigative Reporting, and the Authors Guild did not immediately respond to requests for comment on the newly disclosed figures as the consolidated legal proceedings continue.
Source: The Verge
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment