The Seattle Times Company and Newsday have filed a federal copyright and trademark lawsuit against OpenAI and Microsoft, asking a New York court to order the destruction of AI models trained on their journalism, according to reporting by The Spokesman-Review and the industry press tracker AI Weekly. The complaint, lodged in the Southern District of New York on September 4, adds two of the country’s most storied regional publishers to the growing list of news organizations taking the AI industry to court.
What makes this case different from earlier publisher lawsuits is the remedy being sought. Most copyright plaintiffs ask for damages and licensing arrangements. The Seattle Times and Newsday are asking for something more drastic: that models trained on their copyrighted work be destroyed or, failing that, that the court impose terms that make continued training economically impossible. It is an unusually aggressive ask, and legal observers say it is designed to test how far courts will go once infringement is established.
A pattern of news-industry litigation
The filing joins a crowded docket. The New York Times sued OpenAI and Microsoft in late 2023, and since then a stream of newspapers, wire services, magazines and individual authors has followed. The New York Post and Daily News filed their own suit earlier this year. Teachers and students affected by a mass shooting in Canada have also filed dozens of claims against OpenAI, though those cases concern product safety rather than copyright, and they show how broad the legal front against the company has become.
For the defendants, the timing is awkward. OpenAI has been rolling out GPT-6 Astra, its most capable model to date, and courting enterprise customers who want assurance that training data is clean. Microsoft, which has invested billions in the company and embeds its models across Office products, is named in nearly every publisher case because the copies at issue appear in products both companies sell.
Why regional papers matter
Most of the publicity has gone to national outlets, but regional publishers may have the stronger claims in some respects. Local journalism is expensive to produce and hard to replicate, and news aggregators and chatbots that summarize local reporting strip away the audience those stories depend on. The Seattle Times, an independent, family-owned paper, has been unusually vocal about the survival of local news as a market failure.
Newsday, serving Long Island, brings a similar profile. Both publishers rely on subscription and advertising revenue tied directly to their websites, so any product that answers readers’ questions without sending traffic to the source hits them where it hurts. That economic argument, rather than the legal theory alone, is what the plaintiffs are expected to emphasize at trial.
There is also a trademark component, which is less common in these cases. The publishers allege that AI answers citing or mimicking their mastheads create confusion about the source of information, and that the value of a news brand depends on readers knowing where a story came from. Trademark claims survive on different grounds than copyright ones, so including them gives the plaintiffs a second, independent path to relief if the copyright theory stumbles.
The two plaintiffs also bring complementary strengths. The Seattle Times owns additional Washington-state publications and a regional wire operation, which multiplies the number of registered works at issue. Newsday is one of the largest suburban dailies in the country with a deep archive going back decades. A larger, better-documented corpus makes it harder for the defense to argue the material was marginal to training, an argument that has worked in other contexts.
What happens next
OpenAI has consistently argued that training on publicly available text is fair use, and courts have not yet definitively settled the question. Early rulings in related cases have gone both ways on procedural issues, and no US court has yet ruled on the core question of whether model training itself infringes copyright. That means the Seattle Times case could take years, or settle first, as several earlier publisher suits did, with licensing deals that pay for the past and cover the future.
The destruction remedy is the wildcard. Courts can order infringing materials destroyed under the Copyright Act, but applying that to a model with billions of parameters trained on a mixture of licensed, public and disputed data is uncharted. Legal scholars have debated whether a model can even be considered a “copy” of anything. If the judge lets the claim proceed, the case could become the first real test of that theory, with consequences for every AI company that has scraped the web.
Discovery will be brutal either way. Publisher cases force OpenAI to open up training datasets, ingestion logs and internal communications about what the company knew about copyright exposure and when. Documents from the New York Times case, filed under seal, have already leaked in fragments and shown employees discussing data practices in candid terms. Every additional plaintiff widens that record.
For now, the practical effect is pressure. Every new publisher suit makes licensing look cheaper than litigation, and OpenAI has already signed content deals with several major outlets, including the Associated Press, News Corp and the Financial Times. Those deals cover the signatories, not the industry, and each plaintiff that refuses to settle keeps the legal risk alive. The Seattle Times and Newsday are betting that the courts, not the market, should set the price, and that a destruction order, however unlikely, is the leverage that gets everyone else to the table.
discussion