Microsoft disputes copyright infringement claims in AI chatbot training

The company’s defense centers on data derived from millions of user interactions specifically selected to trigger matches with news content. Out of 8.2 million logs, Microsoft identified only 59,545 instances containing at least 16 words shared with source material. Even more restrictive metrics showed that an expert for the authors' guild found a mere 24 responses containing 30 or more matching words across the entire dataset. Microsoft asserts these figures demonstrate that the AI does not function as a substitute for original journalism or literature.

Publishers and authors contend that their works were used to train models that now compete directly with them. They argue that the AI models regurgitate protected text, violating copyright protections. Microsoft maintains that the training process constitutes fair use, as the final products serve purposes fundamentally different from the original works. With the cases consolidated under one judge, the outcome of this summary judgment request could determine whether the litigation proceeds to a full trial or concludes at this early stage.