Microsoft told its employees to keep “secret” a website-scraping project that gathered AI training data for OpenAI, according to news publishers suing both companies. The claim comes from a newly unsealed section of the publishers’ motion for summary judgment, reported by Bloomberg Law on 16 October. These are allegations made by one side in the case, not findings by the court. Microsoft and OpenAI have argued in court that training AI on copyrighted works is fair use.
What the newly unsealed section alleges
According to Bloomberg Law’s account of the filing, OpenAI paid Microsoft to develop and operate a crawler to copy webpage content under the code name “Project Mango”. That content “was then used to train models including GPT-4o”, the publishers say. Bloomberg reports that Microsoft “ordered its employees to keep ‘secret'” the project.
The data “allegedly included copies of more than 160,000 of the publishers’ works”, Bloomberg Law reports. How much OpenAI paid Microsoft, and how many webpages were scraped, remain under seal.
The case is In re OpenAI, Inc. Copyright Infringement Litigation, No. 1:25-md-03143, a multi-district case in the US District Court for the Southern District of New York. The publishers suing include The New York Times.
What is new, and what was already known
Project Mango itself is not new. The name first appeared when an earlier version of the same motion was unsealed on 17 September. At the time, TechCrunch quoted the filing as saying: “Microsoft similarly provided training data to OpenAI through initiatives called Project Taxi and Project Mango.” TechCrunch reported that the companies allegedly assembled the Project Mango data into a dataset containing copies of “at least 160,903 unique works” from the publishers.
The new details from the 16 October unsealing, according to Bloomberg Law, are:
- the alleged instruction to Microsoft staff to keep the project “secret”;
- the allegation that OpenAI paid Microsoft to build and run the crawler;
- the allegation that the data was used to train models including GPT-4o.
The filing also describes what Bloomberg calls the companies’ self-described “horse trading” deals, including that “OpenAI gave Microsoft its full set of GPT-3 training data”. That point had already appeared in the September unsealing. TechCrunch quoted the filing then as saying OpenAI “delivered the entire GPT-3 training dataset to Microsoft”.
The 47 million prompts estimate
The filing also shows that the publishers estimate ChatGPT receives 47 million prompts a day asking for news content, Bloomberg Law reports. The figure is the publishers’ own calculation. They took a study, cited by an OpenAI expert witness, which found that 1.88% of all ChatGPT messages in July 2025 sought news. They then applied that share to the 2.5 billion prompts ChatGPT receives each day.
What remains sealed
Microsoft originally argued that the material should stay redacted. It withdrew some of those requests on 9 October after the publishers objected, according to Bloomberg Law. Still under seal are how many AI training licensing deals Microsoft and OpenAI have struck and how much they paid, quotes from Microsoft internal documents, and parts of OpenAI’s Slack discussions.
Bloomberg Industry Group, which runs Bloomberg Law, has asked to join the case for the limited purpose of opposing the sealing and redaction of certain documents.
Where the case stands
Both sides filed cross motions for summary judgment in September. The judge’s rulings on those motions will decide which claims about AI training practices go to trial, Bloomberg Law reports. In separate motions filed on 4 September, Microsoft and OpenAI asked the court to find that using copyrighted material to develop their models is fair use, and that their models’ outputs transform the material enough to protect them from liability.
The September unsealing drew attention for a quote from Microsoft’s Director of Applied Science, Brent Hecht, who said copying millions of news articles without permission could be the “largest theft of labor in human history”. The publishers’ filing presents such quotes as evidence for their case. How much weight they carry is for the court to decide.
Bloomberg Law’s report does not include a response from Microsoft or OpenAI to the new details, and TSN has not seen one.
What we don’t know
- What the court will make of the publishers’ allegations.
- How much OpenAI paid Microsoft for Project Mango, and how many pages were crawled.
- The exact wording and context of the “secret” instruction, which TSN has seen only as reported by Bloomberg Law.
What this means
- A newly unsealed filing alleges Microsoft told staff to keep Project Mango “secret” and that OpenAI paid Microsoft to run the crawler, whose data was used to train models including GPT-4o.
- These are the news publishers’ allegations in a summary-judgment motion. No court has ruled on them.
- Key figures, including the payment and the number of pages scraped, are still sealed. The coming rulings will decide what goes to trial.
Sources
- Bloomberg Law: Microsoft told staff to keep OpenAI scraping project ‘secret’
- TechCrunch: Microsoft exec called AI scraping ‘the largest theft of labor in human history,’ new unredacted filings reveal
- Bloomberg Law: Microsoft, OpenAI urge court to find AI training is fair use
Related stories
- USA Today and 13 Sister Companies Sue OpenAI: What the Complaint Alleges
- The New York Times, The Economist and AFP Join More Than 40 Publishers in SPUR, a Coalition on AI Usage Rights
- Not Fair Use: What the Third Circuit Just Said About Training AI on Westlaw Headnotes
- Meta’s New Commercial Terms Let It Crawl Sites Linked to Business Accounts and Use the Content to “Train” Its Products
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X
