OpenAI AI training called 'astonishing theft' in unsealed documents

Microsoft employees raised concerns over OpenAI’s use of millions of news articles and other copyrighted material to train its artificial intelligence models, with one internal document describing it as the potential “largest theft of labor in human history,” according to The New York Times, which cited newly unsealed court documents.

The documents, unsealed Thursday, show Microsoft Director of Applied Science Brent Hecht describing the practice as an “astonishing theft of unprecedented proportions” and warning that it could create a “doom loop” that would ultimately undermine the quality of large language models.

Hecht also reportedly warned that OpenAI may have engaged in an “accidental cover-up” while attempting to identify content from The New York Times and other publishers that were involved in the lawsuit.

Microsoft spokesman Alex Haurek said Hecht’s internal memos did not represent the company’s views.

“Millions of people around the world will soon consider large models ‘hoovering up’ all their work to be an astonishing theft of unprecedented proportions,” the 2023 document said.

'Existential threat' to publishers

OpenAI executives also expressed concern that ChatGPT could divert users from traditional news sources.

Nick Turley, who led the team developing ChatGPT, wrote in a June 2023 memo that AI posed an “existential threat” to publishers. In a February 2024 memo, he said AI products “will get more and more substitutive as they get better.”

An OpenAI engineer similarly acknowledged in a court document that “no matter how prominently we show the links, users won’t click.”

OpenAI has signed licensing agreements with numerous news publishers since ChatGPT was launched in 2022, but concerns persist over whether generative AI could reduce traffic to news websites.

Court documents also show OpenAI President Greg Brockman responding “ah nice” after a staffer mentioned developing “a hack” to bypass The New York Times paywall.

Microsoft CEO Satya Nadella said in a deposition that paywalled content “should be licensed,” adding that he would have required OpenAI to retrain its models if he had known they were trained on paywalled material.

Copyright dispute

The case for intellectual property violation was filed by The New York Times against OpenAI and Microsoft in late 2023, and 11 other publishers later joined the lawsuit.

US District Judge Sidney H. Stein is considering motions for summary judgment, while additional documents are gradually being unsealed.

The publishers accuse OpenAI and Microsoft of violating copyright law by using millions of articles to train AI systems without permission or payment.

The companies argue that their use of the material is protected by “fair use,” maintaining that AI transforms the content into new works rather than simply reproducing or replacing the original articles.