Stories
NYT Lawsuit: OpenAI, Microsoft Scraped News for AI Training, Feared "Doom Loop"
- •Newly unredacted court filings in The New York Times' copyright dispute with OpenAI and Microsoft reveal extensive scraping of news content for AI training, including paywalled material.
- •Internal documents show concerns within Microsoft about generative AI creating a "doom loop" for publishers, with Copilot reducing traffic to The Times' website by up to 93%.
- •Microsoft CEO Satya Nadella testified that paywalled material should be licensed for AI training and would have considered retraining if aware of OpenAI's use of such content.
- •OpenAI's Nick Turley reportedly described ChatGPT as an "existential threat" to publishers, with its products becoming increasingly substitutive.
- •The Times alleges massive scale, with OpenAI's datasets containing millions of documents from nytimes.com and other publishers, and claims of removing copyright notices during data preparation.
Key Questions Answered
Newly unredacted court filings provided a closer look at how news material was gathered for AI training and highlighted internal concerns regarding the commercial consequences for publishers. The documents detailed large-scale collections of news articles, the use of web indexes, and allegations that researchers found ways to access paywalled content.
According to a Microsoft analysis cited in the filing, click-through rates to The New York Times website fell by as many as 93 percent when compared with conventional Bing search. A January 2024 presentation by Microsoft applied the term 'doom loop' to this drop, warning that reduced traffic could ultimately affect both publishers and the AI systems depending on their material.
According to the court filing, Satya Nadella stated in a deposition that paywalled material should be licensed when used for training or grounding. He further testified that had he known OpenAI was training on paywalled content, he would have considered exercising Microsoft's contractual rights to require retraining.
