On September 5, 2026, Beijing time, The Seattle Times Company and Newsday LLC sued nine OpenAI-related entities and Microsoft Corporation in the U.S. Federal District Court for the Southern District of New York. The complaint filed on September 4 alleges that the defendants copied the copyrighted news content of the two newspapers without permission or compensation for use in training, fine-tuning and running generative artificial intelligence products. Case number 1:26-cv-07644, and the court has not yet ruled on any of the substantive disputes.
From crawling charges to retrieval generation
The 38-page complaint lists seven claims, including direct and vicarious copyright infringement, two DMCA claims involving the removal or distribution of copyright management information, and federal, Washington state and New York state trademark dilution claims. The plaintiff seeks statutory damages or the option of actual damages and profits of the defendant, as well as a permanent injunction, seizure or destruction of copies of its work, training data sets and large language models, and a jury trial. The above contents are all litigation requests and have not been approved by the court.
The plaintiffs allege that the defendants obtained content from two newspaper websites through automated scraping, including articles behind paywalls, and then used copies to build training data sets, train and fine-tune models. The complaint also includes ChatGPT, Copilot and Bing-related functions within the scope of the disputed products, claiming that these services can output verbatim text or similar content of the plaintiff’s works. The complaint does not attach the defendant's formal defense to this case, so at this stage it can only be confirmed that the charges have been filed, but it cannot be confirmed that the conduct described has been judicially recognized.
Model testing and newspaper industry interests
The complaint describes separately the data copying during the training phase and the retrieval enhancement generation after the product is launched. According to the plaintiff, the training data allegedly involves WebText, WebText2, Common Crawl and Bing search indexes; search enhancement generation will search the storage index or the real-time Internet after the user raises a query, add search text prompts, and then generate answers. The data source, crawling path and specific usage methods are all narratives proposed by the plaintiff to explain the theory of infringement. No independent technical evaluation was obtained in this round.
The plaintiff also submitted a ChatGPT output test, saying that when the prompt only contained the article title, publication date and URL, the model could still give verbatim or approximate paragraphs of the news involved. One example involves the Seattle Times' reporting on the Boeing 737 MAX accident and subsequent federal investigation; the complaint alleges that the model outputted 88 English words in a row. This number comes from test materials presented by the plaintiffs, which have not yet been reviewed by the court, and no independent technical agency has reproduced the test conditions and results in this round of materials.
Copyright management information issues
In the complaint, the two plaintiffs list subscriptions, advertising and content licensing as the main business foundations that support news production, and argue that generative AI output will reduce the need for readers to visit their websites and weaken the subscription and licensing market. The Seattle Times' principal place of business is in Seattle, Washington, and Newsday's principal place of business is in Melville, New York; the lawsuit states that the two newspapers have won a total of 30 Pulitzer Prizes, 11 and 19 respectively. The identities and award numbers come from court documents, and market harm is still a claim the plaintiffs have yet to prove.
On the issue of copyright management information, the plaintiffs allege that the content extraction process removes or alters article titles, author names, copyright notices, and terms of use, and alleges that OpenAI and Microsoft shared copies with each other that removed such information. The two newspaper companies also objected to the use being considered fair use on the grounds that model training and output services compete with them for readers, advertising and subscription revenue. Whether the copyright management information has been illegally deleted and whether the related use constitutes fair use will need to be determined by the defendant's defense and the court's trial.
Media Validation and Program Boundaries
A public stream of information published by TechCrunch on September 5 provides independent cross-confirmation: two news organizations have sued OpenAI and Microsoft, the dispute involves news content allegedly used to train artificial intelligence. The public summary does not further confirm the source of the training data, the crawling method, the 88-word test or the specific damage, and does not include the defendant's response. Therefore, the technical details and legal claims in the report must be clearly attributed to the formal complaint, and the media’s identification of the subject of the incident cannot be expanded into an endorsement of all accusations.
The only progress in the proceedings that can be confirmed so far is that the complaint has been filed. The public case materials reviewed do not include formal defendant defenses, court schedules, case consolidation decisions, or substantive adjudication, nor do they indicate that the court has approved requests to destroy training data sets or models. Subsequent cases should be based on the case files of the Federal District Court for the Southern District of New York; before there is a defense, evidence review or adjudication, claims such as grabbing without permission, model preservation of works, market substitution and intentional infringement are still the plaintiff's statements and cannot be regarded as established legal conclusions.