MICROSOFT: COPILOT BARELY USED NYT CONTENT

In defense against copyright infringement claims, Microsoft submitted 8.2 million Copilot chat logs as evidence that the AI system does not substantially copy protected content. The company argues that the logs—selected because they involved keywords related to copyrighted material—demonstrate Copilot's minimal reproduction of news articles and published works. Microsoft contends the chatbot rarely outputs even complete sentences from original sources, let alone substantial passages that could serve as substitutes for purchasing or accessing the original content. The disclosure comes as part of the lawsuit's discovery process, where evidence is exchanged between parties. News publishers and authors, including The New York Times, are seeking damages over claims that Microsoft used their copyrighted material to train Copilot without permission or compensation. Microsoft's argument hinges on distinguishing between training data and output—asserting that while Copilot may have been trained on published works, the chatbot does not simply regurgitate that content to users. The company suggests this distinction is legally significant, as copyright claims typically require demonstration of substantial copying in the final product. The case represents a broader tension in AI development. Publishers and authors contend that training large language models on copyrighted material without licensing constitutes infringement. Tech companies counter that such training falls under fair use, a legal doctrine permitting limited use of copyrighted material for transformative purposes. Microsoft's filing strategy suggests the company believes proving minimal output reproduction strengthens its fair use defense. However, publishers' legal teams will likely scrutinize the methodology used to select the 8.2 million chat logs, questioning whether they represent typical Copilot usage patterns or a carefully curated subset. The case continues through the court system, with both sides building their arguments around how AI systems should legally interact with copyrighted material.