Seattle Times and Newsday Sue OpenAI and Microsoft: An 88-Word Reproduction Test Goes to Court

Whether this lawsuit changes the cost of using journalism in generative AI depends on whether the publishers can connect particular model outputs to specific registered works, rather than merely show that a crawler could reach their websites. Over the next three to six months, an order requiring article-by-article allegations, identified model versions, and reproducible output conditions would move the dispute toward testable evidence. An early dismissal of those central claims would sharply reduce its immediate effect on licensing negotiations. The Seattle Times and Newsday filed the case against OpenAI and Microsoft in the Southern District of New York on 2026-09-04; the docket number is 1:26-cv-07644.

The 38-page complaint alleges that the companies copied paywalled articles without permission and used them in training datasets, fine-tuning, and retrieval-augmented generation. It further alleges that titles, bylines, copyright notices, and other copyright-management information were removed. These statements are allegations by the plaintiffs, and the court has not found either defendant liable. The complaint pleads seven counts, including direct and vicarious copyright infringement, two claims under the Digital Millennium Copyright Act, and trademark-dilution claims under federal, Washington, and New York law.

The publishers place model reproduction at the center of their evidence. According to the complaint, a test supplied only an article headline, date, and URL, after which a model reproduced 88 consecutive words from a Seattle Times investigation of the Boeing 737 MAX crisis. It presents further examples involving Newsday text. The test was designed by the plaintiffs, however, and the filing does not disclose a complete prompt log, a reproducible model snapshot, an independent replication, or the size of the tested sample. One example therefore cannot establish how often reproduction occurs or whether every named model retains the same material.

Paywalls, RAG, and a licensing market

The publishers describe two separate technical paths: copying during model training and copying during live retrieval. Newsday says it has operated a complete paywall since August 2022 and that its robots.txt file rejects OpenAI and Common Crawl. The Seattle Times says its terms prohibit using its content for AI training or grounding. The complaint argues that copying for datasets and retrieving articles for RAG answers each infringe its rights. OpenAI and Microsoft may contest the provenance of the data, authorization, fair use, technical causation, and the calculation of damages; the complaint is not a ruling on those defenses.

The economic theory is tied to subscription and advertising revenue. The filing says the two publishers have won 30 Pulitzer Prizes between them, while The Seattle Times has about 108,000 digital-only subscribers and Newsday has roughly 60,000 digital subscribers. The plaintiffs argue that direct answers from AI products reduce referrals, advertising impressions, and subscription conversions. They also point to more than a dozen OpenAI licensing agreements with other media organizations, but they do not establish a negotiated price that these two publishers would have received. A court would still need to separate proof that content was used, proof that an output substituted for an article, and proof of resulting financial loss.

The requested relief includes statutory or actual damages, defendants’ profits, a permanent injunction, and the impoundment or destruction of models and training datasets that incorporate the publishers’ works or derivatives. Those are requests in a complaint, not court orders. TechCrunch reported that Microsoft told GeekWire it was surprised by the lawsuit and remained willing to discuss a solution; the report did not include a substantive OpenAI response. The measurable next steps are whether the case joins the existing multidistrict OpenAI copyright litigation, when the defendants answer or move to dismiss, and whether discovery reaches model versions, training records, and prompt logs.

Sources:

• U.S. District Court for the Southern District of New York complaint: The Seattle Times Company and Newsday LLC v. OpenAI et al.
• TechCrunch: Seattle Times and Newsday are the latest publications to sue OpenAI and Microsoft
• GeekWire: Seattle Times sues Microsoft and OpenAI, alleging they trained their AI on its journalism