Microsoft filings show Copilot rarely reproduces news content.
Microsoft has submitted new legal filings in its ongoing copyright dispute with major publishers and authors, presenting chat log data from its Copilot AI assistant to argue that the service rarely reproduces protected content. As part of the discovery process, Microsoft provided 8.2 million Copilot conversations, specifically selected for keywords related to the plaintiffs' publications. An independent expert hired by news publishers found that approximately 59,545 of these logs contained at least sixteen words matching news content used to train the underlying large language models. However, Microsoft emphasizes that the actual output rarely mirrored full sentences or substantive passages. The publisher-contracted analyst identified only 51 instances of substantial overlap with Center for Investigative Reporting material. In parallel, an expert reviewing author-related claims found merely twenty-four responses containing thirty or more matching words, with matches appearing in just ten of 212 evaluated books. The New York Times, which is actively litigating against Microsoft and OpenAI, rejected the company's characterization. Lead counsel Ian Crosby stated that discovery materials indicate Microsoft and OpenAI appropriated journalistic work to build commercial products that directly substitute for the original reporting, threatening newsroom sustainability. Representatives for the Center for Investigative Reporting and the Authors Guild have not yet commented on the filings. Microsoft's data release underscores its broader legal strategy, which frames the use of copyrighted material for AI training as protected fair use. The company maintains that while large language models ingest published works, their downstream applications serve fundamentally different purposes. The occasional textual overlap, Microsoft argues, does not negate the transformative nature of AI training or undermine the utility of the resulting systems. The filings arrive as Microsoft presses the presiding judge to grant summary judgment, which would terminate the consolidated lawsuits at an early stage. The cases involving news publishers and book authors have been merged under a single federal judge to streamline proceedings, despite initial objections from the plaintiff groups. Adding to the procedural landscape, the Trump administration recently filed a statement of interest in the New York Times lawsuit, lending administrative support to OpenAI's position. If the court denies the motion for summary judgment, the copyright litigation will proceed to trial, setting a potentially defining precedent for how artificial intelligence systems can legally ingest and utilize copyrighted creative works.
