Microsoft's secret Project Mango scraped over 160,000 works from publishers, including the New York Times, for OpenAI's AI training data, according to unsealed court documents.
TL;DR
- Microsoft's Project Mango scraped over 160,000 publisher works for OpenAI, including the New York Times.
- The project was kept secret, with employees ordered to maintain confidentiality.
- The publishers estimate ChatGPT fields 47 million daily prompts seeking news content.
What happened
Microsoft's Project Mango, a website-scraping initiative, harvested AI training data for OpenAI, including over 160,000 works from publishers like the New York Times. The project was code-named and kept secret, with employees instructed to maintain confidentiality. The data was used to train models like GPT-4o, according to a court document unsealed on Friday.
The publishers suing Microsoft and OpenAI for copyright infringement estimate that ChatGPT fields 47 million prompts a day seeking news content. They cite an OpenAI expert witness's assertion of a study finding 1.88% of all ChatGPT messages in July 2025 sought news, applied to the 2.5 billion daily prompts the chatbot receives.
The initial unsealing of the motion last month revealed that OpenAI and Microsoft executives admitted their AI tools were substitutes for articles published by the Times and other outlets, as well as numerous books. Microsoft's Director of Applied Science, Dr. Brent Hecht, was quoted saying the copying of millions of news articles without publishers' permission could be the 'largest theft of labor in human history.'
Why it matters
This revelation highlights the ongoing tension between AI companies and content creators over data usage and copyright infringement. The case could set a precedent for how AI companies can use copyrighted material for training data.
For developers and startups, this case underscores the importance of understanding and navigating the legal landscape around AI training data. It also highlights the potential risks and ethical considerations involved in data scraping and usage.
Investors should pay close attention to the legal and ethical implications of AI training practices. The outcome of this case could have significant implications for the valuation and operations of AI companies, particularly those involved in large-scale data scraping.
Key facts
- Project Mango was a secret website-scraping initiative by Microsoft for OpenAI.
- The project scraped over 160,000 works from publishers, including the New York Times.
- The data was used to train models like GPT-4o.
- ChatGPT fields an estimated 47 million daily prompts seeking news content.
- The case is In Re OpenAI, Inc. Copyright Infringement Litig., S.D.N.Y., No. 1:25-md-03143.
- The motion was partially unsealed on October 16, 2026.
- Microsoft and OpenAI have filed cross motions for summary judgment, which will determine what claims related to AI training practices will head to trial.
Context
This case is part of a broader conversation about the ethical and legal implications of AI training practices. As AI models become more advanced, the need for large amounts of training data has increased, leading to conflicts with content creators and publishers.
The outcome of this case could have significant implications for the AI industry, particularly for companies involved in large-scale data scraping. It could set a precedent for how AI companies can use copyrighted material for training data, affecting the development and deployment of AI models.
The case also highlights the importance of transparency and ethical considerations in AI development. As AI becomes more integrated into our daily lives, it is crucial that companies prioritize ethical practices and respect the rights of content creators.
