Microsoft has told a US court that its Copilot chatbot almost never reproduces copyrighted text — arguing the data shows training AI on news articles and books is fair use. The filings, made on Friday as Microsoft seeks summary judgment in the consolidated copyright cases brought by The New York Times, other news publishers and book authors, draw on 8.2 million Copilot chat logs handed over during discovery.
According to Microsoft, only 59,545 of those conversations contained at least 16 words matching news content used to ground the model — logs the company says it deliberately selected as the ones most likely to involve publishers' content. An expert for the Center for Investigative Reporting found 51 instances of substantial overlap with CIR work in the dataset, Microsoft says, and in the authors' case only 24 responses contained 30 or more matching words. Just 10 of the 212 books evaluated showed any matches at all.
Microsoft argues the numbers hardly undermine the transformative purpose of LLM training, and that systems like Copilot use copyrighted material for purposes very different from the originals. Plaintiffs including the Times, CIR and the Authors Guild declined to comment. The cases were consolidated under a single judge despite publisher objections, and the Trump administration this week filed a statement of interest in the Times case supporting OpenAI, Microsoft's partner in the litigation.
If the judge agrees with Microsoft, the cases could end at an early stage; if not, the copyright fight over generative AI will head toward trial.
