OpenAI paid Microsoft to scrape news in secret. Meta's fine print prices it at zero
A filing unsealed on October 16, 2026, shows OpenAI paid Microsoft to run a "secret" news crawler that copied more than 160,000 publisher works to train GPT-4o, with the payment and page counts still sealed. Meta's rewritten business terms, effective October 30, authorize the same harvest of news sites with no opt-out.
Vincent Jiang · 3 min read
OpenAI$1.18T — OpenAI, private, latest valuation $1.18T paid Microsoft−2.35% — Microsoft, down 2.35 percent today to build and run a website crawler code-named Project Mango, and Microsoft told its staff to keep the project "secret," a court filing partly unsealed on October 16, 2026, shows 1.
The motion comes from news publishers led by the New York Times, who are suing OpenAI and Microsoft for copyright infringement in Manhattan federal court 1. Mango's copies helped train GPT-4o and include more than 160,000 of the plaintiffs' works 1. What OpenAI paid Microsoft, how many pages were scraped and how many licensing deals the companies struck all remain sealed; Microsoft withdrew some redaction requests on October 9 after the publishers objected 1.
Microsoft's own people called the pipeline theft
Brent Hecht, Microsoft's director of applied science, wrote that people would soon consider models "hoovering up" their work "an astonishing theft of unprecedented proportions" and "the largest theft of labor in human history" 2. Nick Turley, OpenAI's head of ChatGPT, said the tools "are largely substitutive, period," and a Microsoft policy document read "LLMs are a product that destroys its supply chain" 2. Microsoft says those words are one employee's view, not the company's 2.
The filing also sizes the demand behind the fight. An OpenAI expert witness found that 1.88 percent of ChatGPT messages in July 2025 sought news; the publishers applied that share to 2.5 billion daily prompts and reached about 47 million news requests a day 1.
Meta prices the same content at a clause
Meta's−4.28% — Meta, down 4.28 percent today rewritten commercial terms take effect on October 30, 2026, and authorize scrapers, crawlers and even "test credentials" to collect content from websites "associated with" business accounts and use it to "train" Meta products 34. There is no opt-out; the only exit is to stop all business use of Facebook, Instagram and WhatsApp, and virtually every news organization keeps such an account 3.
Meta says "there is no change to how Meta uses web crawlers" and that robots.txt, the file sites use to bar crawlers, still governs 35. Three legal experts who reviewed the terms said they could be read as permission to train on news sites, material that "would likely cost many millions of dollars in a licensing market" 3.
Publishers have spent recent years pricing their archives through individual licensing deals; a clause that reaches the same content without payment weakens every one of those negotiations 6.
A judge now holds the price list
The Justice Department has told the court that training on copyrighted works is, by itself, fair use, backing OpenAI and Microsoft against the publishers 7. The publishers' strongest exhibits are the defendants' own words: Satya Nadella, Microsoft's chief executive, testified that paywalled content should be licensed, and internal metrics showed click-through rates to news sites falling 83 to 93 percent when Copilot, Microsoft's AI assistant, answered instead of search 8.
Cross-motions for summary judgment, filed in September, will decide which claims go to trial, and Bloomberg Law, the legal news service asking the court to lift the seals, is fighting the remaining redactions 1. Two dates now matter: the ruling on those motions, and October 30, the day a publisher that keeps using Meta for business is treated as accepting the new terms.
Deepdive
AI-generated from this story and its cited sources. Not investment advice.
