Seattle Times and Newsday Sue OpenAI and Microsoft Over AI Copying

The Seattle Times Company and Newsday LLC filed a federal complaint on September 4 against nine OpenAI entities and Microsoft Corporation, alleging that the companies copied their journalism without permission to train and operate ChatGPT, Copilot and related AI services. The case matters because it challenges alleged copying at several stages, from collecting paywalled articles to generating answers the publishers say can replace visits to their sites, while seeking relief that could reach models and training datasets.

The 38-page complaint opened case 1:26-cv-07644 in the U.S. District Court for the Southern District of New York. It pleads seven causes of action and demands a jury trial. At this stage, the document states the two newspaper publishers’ allegations; it is not a ruling on how the companies obtained the articles, how their systems used them, whether any use was fair, or whether the publishers suffered legally compensable harm.

How the Alleged Copying Chain Works

The complaint starts with acquisition. The publishers allege that automated systems copied material from their websites, including articles behind hard or metered paywalls and pages covered by terms or crawler instructions. They say those copies entered datasets assembled from sources such as Common Crawl and Microsoft’s Bing search index, then passed through repeated model training or fine-tuning.

Retrieval-augmented generation is a different part of the alleged chain. Rather than relying only on what a model processed during training, a search-enabled product can fetch text from a stored index or the live web after training and use it to ground an answer. The complaint alleges that ChatGPT search features and Microsoft products including Copilot and Bing used this route as well. An article retrieved for an answer therefore would not, by itself, show that the same article was included in training data.

The filing ties those inputs to concrete outputs. It says a prompt containing the headline and URL of a Seattle Times investigation into the Boeing 737 MAX produced 88 consecutive words from the article. A separate side-by-side table presents four Newsday examples involving disaster funding, employment data, a fashion-business profile and labor statistics; the alleged overlap ranges from exact wording to close paraphrase. 

The publishers argue that answers containing or summarizing their reporting can satisfy a reader’s query without a visit, subscription or licensing transaction, reducing traffic, advertising and other revenue. That substitution claim supplies part of their alleged harm and their answer to a future expected fair-use defense.

Seven Counts Divide Copyright, Metadata and Brand Claims

The first count alleges direct copyright infringement. The second alleges vicarious liability against Microsoft and specified OpenAI entities, arguing that they had control over and profited from the systems that allegedly made the copies. Those theories depend on proving the underlying copying and the defendants’ respective roles, not merely on showing that an AI product can discuss the same subjects as the newspapers.

Two more counts invoke the Digital Millennium Copyright Act. Copyright management information is the identifying material attached to a work, such as its title, author, copyright notice or terms. The publishers allege that defendants removed this information during extraction and later distributed copies without it. The statute also requires knowledge-related elements that the complaint alleges but the court has not yet evaluated.

The remaining three counts allege trademark dilution under federal law and the laws of Washington and New York. Their focus is attribution rather than copying alone: the publishers say generated or fabricated material associated with their names can blur or tarnish marks built around editorial accuracy. The complaint offers fewer detailed false-attribution examples than it does reproduction examples.

The publishers ask for copyright and DMCA damages, profits where available, permanent injunctions and trademark-related relief. They also ask the court to impound or destroy copies of their works and any large language models or training datasets that incorporate those works or material derived from them.

Nothing in the complaint orders destruction, finds liability, awards damages or decides fair use. The scope and technical feasibility of any remedy would be questions for later litigation if the publishers first establish the legal and factual basis for relief.

Public Statements Leave the Specific Allegations Unanswered

OpenAI gave a general response on September 4, saying through a spokesperson that its models train on publicly available data and that its approach rests on fair use. The company declined to address this complaint specifically, according to Reuters reporting. The statement did not address the alleged paywall bypass or how specific articles entered datasets, appeared in outputs, lost identifying information or became associated with the publishers’ brands.

Microsoft said it was surprised by the filing, valued its local relationship with The Seattle Times and was willing to discuss solutions, according to reporting that carried the company’s emailed response. The statement did not address the complaint’s Microsoft-specific claim that Bing index data and bespoke crawlers supplied publisher material, or the allegations involving Copilot and Bing answers. Neither public statement is a merits response filed in court.

The Publishers Also Joined an AI Fellowship

Microsoft’s surprise has a notable backdrop. In 2024, The Seattle Times and Newsday were among five metropolitan news organizations named in a $10 million local-news AI initiative managed by the Lenfest Institute and funded by Microsoft and OpenAI. The plan paired grants for two-year fellows with enterprise software credits. Newsday’s announced project involved summarizing public data, while The Seattle Times later developed an advertising-sales prospecting tool using services from the same technology companies. The public descriptions cover grants, credits and business tools. But that collaboration alone does not establish a license to train on the publishers’ journalism, a waiver of legal claims or a contradiction in filing suit.

The complaint joins a wider set of publisher cases over the use of news in generative AI. The closest comparison is The New York Times’ separate 2023 lawsuit against OpenAI and Microsoft, which also disputes training copies, outputs, paywall-protected work and substitution for publisher products.

That older litigation was further along by September 4, 2026, when Microsoft and OpenAI sought summary judgment on fair use. It supplies a developed version of questions that the new complaint raises, but it involves a different record and has produced no final ruling that resolves the Seattle Times and Newsday claims.

The complaint sets no case-specific response date. Under the ordinary federal rules, the clerk issues a properly completed summons, the plaintiffs arrange service or seek a waiver, and each defendant then answers or files a motion under Rule 12.