Internal memos released Thursday in a copyright case show that Microsoft employees privately questioned the legality of training generative AI models on news publications. Director of Applied Science Brent Hecht described the scraping of such data as "an astonishing theft of unprecedented proportions" in a memo unsealed as part of the litigation. The filings indicate top staff at Microsoft and OpenAI warned that the practice made a "complete mockery" of fair use and posed an "existential threat" to publishers.

Internal Microsoft data found that the Copilot AI reduced click-through rates on some news domains by as much as 93% compared to Bing, a phenomenon employees termed the "doom loop." The documents also include allegations that OpenAI used a "paywall hack" to access copyrighted content, which OpenAI co-founder Greg Brockman acknowledged in an internal exchange. Microsoft responded that Hecht's view was one employee's "individual perspective."