Microsoft staff warned of 'astonishing theft' in OpenAI training

Microsoft Director of Applied Science Brent Hecht described OpenAI’s use of millions of copyrighted news articles as an “astonishing theft of unprecedented proportions” in internal memos unsealed Thursday, according to The New York Times. Hecht warned the practice could constitute the “largest theft of labor in human history” and might create a “doom loop” that ultimately degrades the quality of large language models by contaminating training data with proprietary restrictions.

He also raised concerns that OpenAI had engaged in what he termed an “accidental cover-up” while attempting to isolate material from The New York Times and other plaintiffs involved in the litigation. A 2023 memo cautioned that “millions of people around the world will soon consider large models ‘hoovering up’ all their work to be an astonishing theft of unprecedented proportions,” though Microsoft spokesman Alex Haurek told reporters the internal communications represented personal views rather than the company’s official corporate position.

'Existential threat' to news industry

OpenAI executives privately acknowledged that ChatGPT could divert traffic from traditional news outlets, with product lead Nick Turley writing in a June 2023 memo that artificial intelligence posed an “existential threat” to the publishing industry. Turley noted in February 2024 that AI products “will get more and more substitutive as they get better,” while an engineer admitted in sworn testimony that “no matter how prominently we show the links, users won’t click,” highlighting concerns that persist despite licensing agreements OpenAI has signed with numerous major publishers since ChatGPT launched in 2022.

The documents also reveal OpenAI President Greg Brockman responding “ah nice” when a staffer described developing “a hack” to circumvent The New York Times paywall and access restricted articles. Microsoft CEO Satya Nadella stated in a deposition that paywalled content “should be licensed,” adding he would have ordered model retraining had he known about the unauthorized use of restricted material. The New York Times and eleven other publishers accuse the companies of violating copyright law by using millions of articles without permission or payment, while OpenAI and Microsoft argue their use constitutes “fair use” that transforms content into new works. The case, filed in late 2023, is now before US District Judge Sidney H. Stein, who is considering motions for summary judgment as additional documents gradually enter the public record.