Unredacted court documents released Sept. 17 in a legal dispute over copyrighted content reveal a scathing internal view of how AI models are built. A Microsoft Director of applied science characterized the training of large language models as "an astonishing theft of unprecedented proportions" in the newly released filings in a case brought by publishers against Microsoft and OpenAI.

The litigation focuses on whether the companies can legally use journalistic material without payment to develop their systems. Microsoft and OpenAI have not yet responded to requests for comment on the unredacted testimony, and the New York Times declined to comment on the proceeding.