Recommended Articles
Legal & Policy

News Publishers Sharpen Copyright Case Against Microsoft and OpenAI

By Angeled · October 17, 2026 · Legal & Policy
News Publishers Sharpen Copyright Case Against Microsoft and OpenAI
News Publishers Sharpen Copyright Case Against Microsoft and OpenAI

On October 16, 2026, news publishers filed a less-redacted version of their motion for summary judgment in the US copyright case against Microsoft and OpenAI over the use of publishers' content to train artificial intelligence models. The filing — a 92-page redacted brief, the second updated public version of their September 4 motion — pushes the litigation into its decisive phase: the plaintiffs ask the court to rule that the companies infringed and that fair use fails as a matter of law. These are the plaintiffs' allegations; nothing has been decided.

The case and the motion

The news cases — brought by the New York Times, the Daily News (DNP), The Intercept, the Center for Investigative Reporting, and Ziff Davis — are consolidated in a multidistrict litigation in the Southern District of New York (No. 1:25-md-03143). The plaintiffs seek summary judgment that OpenAI and Microsoft infringed by acquiring, training on, grounding on, and trading their articles, and that OpenAI intentionally removed copyright management information.

The substitution argument — with Microsoft's own data

The core of the publishers' case is substitution: AI answers replace visits to the publishers' own sites. According to MLex's reporting on the filing, the brief states:

  • 83–93% drops in click-through rates for the New York Times and DNP domains for Microsoft's Copilot "answer engine" compared with traditional Bing search.
  • 51–94% drops for Ziff Davis domains on the same comparison.

The brief quotes Microsoft's lead counsel on why chatbots divert users: their advantage "is that they are actually designed to answer your question" rather than "just giving you a series of links to things you can go look for yourself and maybe find the information" — that is, they divert users from visiting the publishers who created the copied work.

The plaintiffs frame this as a "doom loop": AI models trained on publisher content then answer users' questions directly, draining the traffic that funds the journalism the models depend on.

The harvesting allegations

The brief levels detailed allegations about how the training data was obtained:

  • Project Taxi: From 2019 to 2022, Microsoft allegedly gave OpenAI a copy of the Bing index under the code name Project Taxi — in effect, selling OpenAI a copy of the search index.
  • Project Mango: Microsoft then allegedly built and ran a crawler for OpenAI under the name Project Mango, telling staff to keep it secret.
  • Paywall circumvention: An OpenAI researcher allegedly told co-founder Greg Brockman about "a hack to get around" the New York Times paywall; Satya Nadella is said to have testified that paywalled content should be licensed.
  • Stripped copyright notices: OpenAI's text extractors allegedly removed copyright notices from over 99% of the asserted articles.

What OpenAI paid for the index and the crawler, the number of pages crawled, and three full pages of the brief (pages 6, 18, and 86) remain redacted. Output and grounding claims are pressed against Microsoft only, pending a sanctions motion against OpenAI.

Why this phase matters

Summary judgment is the inflection point of AI copyright litigation. If the court agrees that fair use fails as a matter of law on these facts, the ruling would reshape the economics of training data across the industry — strengthening publishers' hands in licensing negotiations and raising the cost base for model builders. If the court disagrees, it would mark the strongest judicial endorsement yet of training on copyrighted news content.

Either way, the case is establishing the factual record — click-through data, index deals, crawler operations — that other courts and negotiators will reference for years.

What founders and investors should take away

  1. Training-data provenance is now a diligence item. Investors funding foundation models, agents, and retrieval products should expect questions about data sourcing, licensing, and copyright-management-information handling.
  2. Licensing markets are forming in real time. The publishers' leverage grows with each damaging disclosure; startups that secure clean, licensed data pipelines early are buying optionality.
  3. Product design intersects with liability. Features that substitute for source content — answer engines, grounding, summarization — are precisely what the plaintiffs target. How a product attributes, links, and drives traffic back to sources is becoming a legal design constraint, not just a UX choice.
  4. Watch the redactions. What remains hidden — what was paid for the index, how many pages were crawled — will matter enormously when damages are discussed.

Angeled.com connects founders with investors.

Sources: reported October 16, 2026 — MLex; payauthors.com (allegation grading: plaintiffs' filing).

Read Next