'Destroy the Models': Seattle Times and Newsday Sue OpenAI and Microsoft, Escalating the AI Copyright Wars

On September 4, 2026, two of the most recognizable names in American regional journalism walked into the U.S. District Court for the Southern District of New York and asked a federal judge to do something no court has ever done: order the destruction of AI models.

The Seattle Times Company and Newsday — the Pulitzer-winning daily of Long Island — filed a joint copyright and trademark complaint against OpenAI and Microsoft, alleging the companies “methodically scraped” their journalism, including articles behind paywalls, to train and operate ChatGPT, Microsoft Copilot, and Bing’s AI features. The suit seeks unspecified damages and, most strikingly, court orders requiring the “impoundment and/or destruction” of copies of the works, the training datasets built from them, and any AI models that incorporate them.

What the suit actually alleges

The complaint’s core claims will feel familiar to anyone who has followed the copyright litigation wave that began with the New York Times in late 2023. According to the filing and reporting by the Spokesman-Review and Reuters, the newspapers allege that:

• OpenAI and Microsoft scraped the newspapers’ websites — including paywalled content — and incorporated the articles into datasets used to train and operate commercial AI products.
• The resulting models can reproduce passages from their reporting, closely paraphrase articles, and answer user queries in ways that reduce the need to visit the publishers’ sites or buy subscriptions.
• The companies diluted their trademarks by generating content that falsely attributed fabricated material to the Seattle Times and Newsday — the hallucinated-citation problem that has dogged news-adjacent chatbots since their introduction.

The economics motivating the suit are quantified in the complaint itself. Attorneys for the two outlets cited industry data showing that search referral traffic to mid-sized publishers fell 47% year over year as of December 2025. For regional papers that survived the last two decades of digital disruption on the strength of search-driven subscription funnels, that number is existential — and the filing says so explicitly.

“This was not an easy decision,” Seattle Times President and CEO Alan Fisco wrote in an email to employees on Friday. “However, we feel strongly that we must defend our content — which we spend millions of dollars a year to produce — from being used without our consent or compensation.”

The remedy that turns heads: impoundment

Damages claims are standard fare in copyright cases. What distinguishes this filing — and aligns it with the New York Times’ ongoing suit — is the requested remedy: not just money, but the physical destruction of infringing AI assets. The plaintiffs want copies of their works impounded, training datasets destroyed, and any models trained on those datasets destroyed along with them.

Legally, this is the nuclear option. Copyright’s impoundment provisions are typically invoked against counterfeit handbags and pirated DVD presses, not trillion-dollar software systems. But the ask is strategically coherent: it forces the court to confront what “making whole” means when a copyrighted corpus has been irreversibly baked into model weights. You cannot un-train a model the way you can pulping a book run — and the plaintiffs want the court to say that this irreversibility is precisely the problem.

The defendants respond — politely

Both companies struck conciliatory notes in public while conceding nothing on the merits.

A Microsoft spokesperson said the company was “surprised by the lawsuit” but “appreciate[s] the importance of the Seattle Times to our region” and is “always happy to sit down and explore solutions to this type of dispute.”

OpenAI’s statement leaned on its standard fair-use defense: its models are trained on publicly available data and are “grounded in fair use, which helps hundreds of millions of people improve their daily lives and delivers benefits such as empowering human creativity, science, and medical research.”

The irony in the replies is easy to miss but sharp: OpenAI’s warm words about empowering creativity are aimed at the same newsrooms whose content it stands accused of taking. And Microsoft’s offer to “sit down” comes after years in which licensing negotiations between AI firms and publishers produced some of the industry’s landmark deals — for the likes of the Associated Press, Axel Springer, and News Corp — while regional and mid-sized publishers were largely left negotiating alone, or not at all.

The DOJ just picked the other side

Context matters here, and the timing is brutal for the plaintiffs. Three days before this filing — on Tuesday, September 1 — the U.S. Department of Justice entered the parallel New York Times v. OpenAI case on the side of the AI companies, filing a statement of interest arguing that AI development is a matter of national interest. If a judge found copyright violations, the DOJ argued, it would stifle “scientific progress while hindering American prosperity and economic mobility,” and that the “creative possibilities” and benefits of AI models “far outweigh any competitive harm” from training on publishers’ content.

The executive branch’s intervention reframes what was once a narrow commercial dispute as a question of industrial policy. With China’s open-weight models pressuring American AI firms on cost and reach, Washington has concluded that the training-data pipeline is strategic infrastructure — and it is now willing to say so in court. Every publisher suit filed after that filing, including this one, lands against that backdrop.

A growing docket, and the Anthropic counter-example

The Seattle Times and Newsday are the newest plaintiffs, not the only ones. The complaint joins a docket that now includes:

• The New York Times’ pioneering 2023 suit against OpenAI and Microsoft, still ongoing in the SDNY.
• A 2024 suit by a group of Alden Global Capital-owned daily newspapers.
• A 2025 action by 35 publishers representing more than 400 community newspapers.
• Class actions from authors and musicians spanning multiple circuits.

And there is the counter-example that publishers keep citing: Anthropic’s $1.5 billion settlement with a coalition of authors and publishers last year, resolving claims that Claude was trained on copyrighted books. That deal demonstrated that the industry can price a licensing settlement when it chooses to — which makes the refusal to strike similar deals with news publishers look less like necessity and more like strategy.

The complicated Seattle-Microsoft relationship

One wrinkle specific to this suit: Microsoft and the Seattle Times are not strangers. Microsoft Philanthropies underwrites Seattle Times journalism projects (the paper maintains editorial independence), and the Times was among the publishers that received a Lenfest Institute grant in 2024 to hire a two-year AI fellow — a program funded with $10 million from Microsoft and OpenAI. The newsroom holds some ChatGPT licenses for data analysis and has a public AI policy allowing limited, labeled use of AI tools for transcription and analysis after risk review.

In other words, the plaintiff takes AI-industry money for fellowships, uses ChatGPT in its newsroom, and is nonetheless asking a court to destroy models trained on its archive. That is not hypocrisy — it is a precise statement of the publishers’ actual position: AI is fine; taking their content without consent or compensation is not. Fisco’s internal note drew the same line, saying the lawsuit “is not about impeding AI innovation” but about “making sure that innovation does not come at the expense of our business model.”

What happens next

The case will proceed on the familiar two-track path: a likely motion to dismiss resting on fair use, followed by years of discovery if it survives — with the SDNY’s eventual rulings on the training-data question shaping not just this suit but the entire publisher litigation wave.

The stakes are easiest to see in the remedy. If a court ever seriously entertains impoundment of a frontier model, the settlement math changes overnight — every unlicensed corpus in every model becomes a latent liability, and licensing deals go from PR expense to existential insurance. If instead the fair-use defense holds across the docket, the implicit message to every newsroom is that their archives were a one-time, uncompensated substrate for the AI economy.

Either way, the Seattle Times and Newsday have just made sure the question gets asked in the loudest possible terms. Two regional newsrooms — one on Puget Sound, one on Long Island — are now asking a federal court to weigh the destruction of some of the most valuable software artifacts ever built against the survival of the journalism that helped train them.