Microsoft Claims Copilot Rarely Reproduces Copyrighted Content in NYT Lawsuit

Microsoft has argued in recent legal filings that its AI assistant Copilot rarely reproduces full sentences from news articles or books, asserting that it does not generate substantive portions capable of substituting for the original works. The filings come as part of ongoing copyright infringement lawsuits brought by publishers, including The New York Times, and book authors. During the discovery phase, Microsoft turned over 8.2 million Copilot chat logs to an expert retained by news publishers. The company noted that these logs were selected specifically because they included search terms implicating the plaintiffs' websites. According to Microsoft, the analysis showed that only 59,545 conversations contained at least 16 words matching the news content used to ground the AI model. Additionally, an expert for the Center for Investigative Reporting identified 51 instances of "substantial overlap" with CIR material within the dataset. In the suit involving authors, an expert discovered that only 24 responses out of the 8.2 million conversations shared at least 30 identical words, with matches found in just 10 of the 212 books examined. The New York Times rejected Microsoft’s interpretation. Ian Crosby, lead counsel for The Times, stated that the discovered documents confirm Microsoft and OpenAI took content without authorization to build commercial products that undermine journalism. Microsoft maintains that the limited duplication supports its defense that using copyrighted material for AI training constitutes fair use, as the technology serves a fundamentally transformative purpose.