Microsoft and OpenAI Internal Files Revealed: AI Training on News Content Called Massive Theft

The New York Times and other news organizations are pushing forward with lawsuits against Microsoft and OpenAI over training data copyright. Previously confidential internal files have now been made public. Brent Hecht, head of Microsoft's applied science division, warned in an internal document that using news content for AI training is 'unprecedented mass theft' and possibly 'the largest labor theft in human history.' He also questioned Microsoft and OpenAI's claim that AI training on news content falls under fair use.

Microsoft internal files raised a 'doom loop' risk: AI products keep reducing news site traffic and revenue, causing news organizations to cut content production, and AI models then lose reliable information sources. Microsoft data shows some sued news outlets saw click-through rates drop 83% to 93%, while others fell 51% to 94%. Microsoft CEO Satya Nadella admitted in testimony that chatbots providing information directly on AI platforms amounts to 'taking clicks away from news sites.'

Similar views exist inside OpenAI. ChatGPT head Nick Turley confirmed internally that training commercial AI products on news content could pose an 'existential threat' to publishers, saying users have 'little reason' to click original links. News organizations also allege that internal staff knew AI products could output content highly similar to training data. Microsoft responded that its AI products are transformative fair use, and Hecht's document represents only one employee's personal view.