Microsoft Official Privately Called AI Training on News Content "Theft," Court Filings Reveal

Freshly unredacted court documents from a three-year-old copyright dispute have brought to light damaging internal communications from both Microsoft and OpenAI executives, acknowledging that their approach to gathering training data crossed ethical and legal lines.

The Damning Admissions#

Court filings in The New York Times' lawsuit against OpenAI and Microsoft now reveal that senior leaders at both companies privately recognized the problematic nature of their practices. A top Microsoft executive characterized the data acquisition strategy as "the largest theft of labor in human history," while OpenAI's own ChatGPT team flagged the technology as creating an "existential threat" to the news organizations whose content had trained these systems.

Microsoft CEO Satya Nadella went further during a deposition, stating unambiguously that any paywalled material should require licensing for use in AI development. He indicated he would have demanded that OpenAI retrain its models had he known the company had scraped content behind paywalls.

How the Companies Obtained Content#

The documents detail the scale and sophistication of the data-collection operation. OpenAI and Microsoft coordinated through initiatives known as Project Taxi and Project Mango to systematically gather copyrighted material. The filings reveal that OpenAI's training datasets alone contained over 91,000 copies of articles from The New York Times, Daily News, and other publishers. One dataset derived from Common Crawl—a free web archive—included more than 2 million documents from nytimes.com.

To maximize their haul, company researchers developed strategies to bypass subscription barriers without triggering detection. When an OpenAI scientist mentioned discovering a "hack to get around nytimes paywall," leadership responded approvingly. The teams also deliberately removed copyright notices from source material before feeding it into their models, explicitly to prevent AI systems from reproducing those notices in their outputs.

The Competitive Threat#

Internal Microsoft analysis underscored the economic harm to publishers. Data showed that Microsoft's Copilot answer engine reduced click-through rates to New York Times content by as much as 93% compared to traditional search results—what one Microsoft director characterized as a "doom loop." A Microsoft presentation from January 2024 framed this dynamic as fundamentally unsustainable, noting it was "highly unusual that an end-product threatens the economic foundations of its essential suppliers."