Unredacted Filings Detail Microsoft Executive's Concerns in AI Copyright Suit

Summary

Newly unredacted filings in a Manhattan federal copyright case brought by the New York Daily News, The New York Times, the Orlando Sentinel and other newspapers against Microsoft and OpenAI disclose internal comments about the use of news content to build AI systems. Brent Hecht, Microsoft’s director of applied science, wrote that millions of people could regard large models absorbing others’ work as “an astonishing theft” and potentially the largest theft of labor in human history, while noting that creators generally did not intend or receive compensation for such use. The newspapers’ lawyers argue that the disclosures show the companies understood their conduct as theft rather than fair use; Microsoft says Hecht was expressing an individual, non-legal view and maintains in its court filings that the uses are transformative and Copilot does not replace journalism. Other Microsoft documents describe a “doom loop” in which AI answers reduce visits to the sources that supply content needed by the models. The newspapers allege that the companies copied billions of web pages and cite more than 3.9 million copies from The Times and 7.3 million from The News and affiliated papers for training. They also seek sanctions against OpenAI over alleged evidence destruction and concealment of its ability to locate copied stories. The filings cite Microsoft-recorded declines of 83% to 93% in clicks to the newspapers’ links, survey findings that former subscribers to The News and affiliated papers were 28.6% more likely to ask AI for news, and data showing 36% of surveyed Times subscribers using ChatGPT said they no longer needed news from The Times. Microsoft experts said at least 1.3% of 45,000 Copilot conversations concerned current affairs, while OpenAI data cited about one million weekly prompts for reliable local news. The technology companies argue that their models transform the material and qualify for fair use; the newspapers contend that outputs can distort reporting and contribute to low-quality AI-generated news.