The Seattle Times and Newsday Sue OpenAI and Microsoft, Adding Trademark Dilution to Copyright Claims

On September 4, 2026, two US newspapers — The Seattle Times Company and Newsday LLC — sued nine OpenAI entities and Microsoft in the US District Court for the Southern District of New York (No. 1:26-cv-07644)¹. The 38-page complaint demands a jury trial.

The complaint brings seven counts, three of which rest on trademark rather than copyright¹. Suits by news organizations against OpenAI and Microsoft have continued since The New York Times filed in 2023².

Copyright, the DMCA, and trademark, in three layers

The complaint brings the following seven counts¹.

The first is copyright infringement (17 U.S.C. §501); the second is vicarious infringement. The vicarious count names Microsoft and some of the OpenAI entities, alleging that Microsoft “provided the massive computing platform used to store, process, and reproduce the training datasets containing Plaintiffs’ works.” The Microsoft commercial products the complaint identifies are Copilot and Bing Chat.

The third and fourth counts are brought under the DMCA (17 U.S.C. §1202(b)(1) and (b)(3)) over the removal of copyright management information (CMI) and the distribution of copies with CMI removed. The complaint describes CMI as including article titles, author names, and copyright notices.

Counts five through seven are trademark dilution claims, under federal law (15 U.S.C. §1125(c)) as well as Washington State law (RCW 19.77.160) and New York General Business Law §360-L. The plaintiffs point to four registered THE SEATTLE TIMES marks (all registered on April 30, 2019) and one registered NEWSDAY mark (registered March 25, 1997), stating that each is incontestable under 15 U.S.C. §1065.

The claim that a headline and a URL returned 88 words

What the complaint sets out are the plaintiffs’ allegations, not findings by the court. The scraping, verbatim-reproduction and RAG allegations described below each carry the express qualifier “upon information and belief.” With that in mind, here is what the plaintiffs specifically allege.

On training data, they allege that the defendants spent years methodically scraping their websites with automated bots, bypassing their paywalls, and “ignoring decades-old norms” to obtain copies of their articles¹. The datasets named are those derived from WebText, WebText2, Common Crawl, and Microsoft’s own Bing search index. For Newsday, the complaint also states that its robots.txt instructed both OpenAI and Common Crawl not to crawl the site.

On output, the plaintiffs allege that the models reproduced substantial passages of their articles verbatim from prompts supplying nothing more than the headline, publication date, and URL¹. The complaint treats this as confirming that “the reproduced text was recalled from the model’s training rather than copied from text a user had supplied to it.” The example it gives is an article from The Seattle Times’ Pulitzer Prize-winning series on the 2019 Boeing 737 MAX crashes, from which it says a model reproduced 88 consecutive words verbatim in response to a simple prompt containing the headline and URL. The complaint adds that this shows “readers determined to evade the Plaintiffs’ paywalls need only ask ChatGPT for help.”

It further alleges ongoing copying through RAG-based search: that the defendants’ GenAI products keep making new copies of current articles, and reproduce the plaintiffs’ reporting by scraping their sites directly in response to user queries about current events.

Framing misattribution as damage to the brand

The trademark dilution counts reach for something copyright does not readily cover.

The complaint states that the defendants’ products “hallucinate” content that they misattribute to the plaintiffs, or falsely attribute the plaintiffs’ content to another party, and that the plaintiffs and their trusted brands are tarnished by that misinformation¹. It frames this as tarnishment under trademark law (§1125(c)(2)(C)).

It also alleges blurring (§1125(c)(2)(B)), on the theory that the conduct “whittles away the exclusive, source-identifying association the public has long held between Plaintiffs’ Marks and Plaintiffs’ journalism, associating those Marks instead with Defendants’ AI-generated substitute content over which Plaintiffs exercise no editorial control.”

In other words, this case puts whose work the output is presented as squarely at issue, not just whether the material could be used for training.

The remedy sought is destruction of the models and datasets

The requested relief is broad. Alongside statutory damages or actual damages plus the defendants’ profits (17 U.S.C. §504), statutory damages for each §1202 violation, and a permanent injunction, the plaintiffs ask under 17 U.S.C. §503 for the impoundment or destruction of all LLMs and training datasets incorporating their works¹. On the trademark counts they also seek treble damages and attorneys’ fees on the basis of bad faith and willful conduct.

The complaint does not state a total damages figure.

The plaintiffs are represented by Klaris Law PLLC. Microsoft’s spokesperson has said the company did not expect the suit and remains willing to discuss a resolution. OpenAI’s spokesperson has said the company trains its models on data that is already public and regards that as covered by fair use³. TechCrunch notes that the filing is particularly striking because Microsoft and OpenAI have funded some of The Seattle Times’ journalism projects and fellowships².

Stepping straight into the output side the DOJ left open

The complaint was filed three days after the US Department of Justice filed a statement of interest in an AI copyright case. That filing argued that training an LLM on copyrighted works is “exceedingly transformative” and qualifies as fair use, while expressly reserving that certain uses at the output stage may not be transformative.

The Seattle Times and Newsday complaint concentrates its claims on exactly that reserved side. Verbatim reproduction, CMI removal, and brand damage from misattribution are all issues that can stand on their own even if training is held lawful. When music publishers sued Anthropic and two of its founders in August, the dispute ran across the acquisition, training, and output of lyrics; here a separate body of law, trademark, is layered on top. Separating the stages is also how Japan’s Agency for Cultural Affairs guidance is organized, treating the training stage apart from generation and use.

For organizations using generative AI at work, nothing changes immediately: a complaint is a set of allegations, not a ruling. What is readable in it is a shift in where the argument sits. The DOJ has staked out a position on the legality of training data, but whether output reproduces existing works — and whose name that output carries — remains unsettled. For any organization that publishes AI output under its own name, misattribution being litigated as a trademark matter is part of the picture too. As Debian drew its own line by passing a general resolution on the responsible use of generative AI, leaving responsibility for submitted work with the contributor, the need to decide for yourself how output gets checked is not going away.

Sources

1. Complaint, The Seattle Times Company v. OpenAI Inc. - The original 38-page complaint filed in the US District Court for the Southern District of New York on September 4, 2026 (No. 1:26-cv-07644), via the CourtListener RECAP archive
2. Seattle Times and Newsday are the latest publications to sue OpenAI and Microsoft - TechCrunch (September 5, 2026)
3. The Seattle Times sues OpenAI, Microsoft over copyright infringement - A Seattle Times staff report by Alex Halverson (September 4, 2026), distributed by The Spokesman-Review