Unsealed motion: OpenAI paid Microsoft to operate secret 'Project Mango' scraper for AI training data

What happened

News publishers in the OpenAI copyright MDL (S.D.N.Y. No. 1:25-md-03143) partly unsealed their motion for summary judgment on 2026-10-16. The filing says OpenAI paid Microsoft to develop and operate a website crawler, code-named "Project Mango," that copied webpage content used to train models including GPT-4o. The dataset allegedly included copies of more than 160,000 of the publishers' works. Microsoft ordered its staff to keep the project "secret."

What's still under seal

How much OpenAI paid Microsoft, the total number of webpages scraped, how many licensing deals the companies have struck and their payment amounts, quotes from Microsoft internal documents, and portions of OpenAI Slack discussions.

Context from the same filing

The publishers also estimate ChatGPT fields 47 million news-seeking prompts per day, and the earlier unsealed portion revealed executives admitting their AI tools substitute for Times and other publishers' articles and numerous books.

All of the above are allegations in the publishers' motion. The parties filed cross-motions for summary judgment in September 2026, which will determine which claims head to trial.