Seattle Times and Newsday Sue OpenAI and Microsoft
Two regional papers filed in the Southern District of New York over training data. The interesting part is not the accusation, it is the remedy they asked for.
The Seattle Times and Newsday have filed a copyright lawsuit against OpenAI and Microsoft, landing at the end of last week in the US District Court for the Southern District of New York. The two papers say their journalism, including work behind paywalls, was scraped and used to train and operate ChatGPT, Microsoft Copilot and Bing's AI features. What makes this filing worth ten minutes of your attention is not the accusation, which is now familiar, but the remedy: the papers are asking a court to order the destruction of the training datasets and models that contain their work.
What the papers actually filed
The complaint describes generative AI as "a snake eating its own tail," a system that consumes the journalism it depends on and then substitutes AI-generated summaries for the original, taking the traffic and the ad revenue with it. The papers argue the industry could end up "broken beyond repair," and describe AI systems as "rapacious consumers, devouring human-authored content" to produce "copies and derivative imitations."
Two details separate this from the pile of similar suits. First, Microsoft and OpenAI have previously funded journalism projects and fellowships at The Seattle Times, so the plaintiff here is not a stranger to the defendants. Second, a Microsoft spokesperson responded by saying the company was surprised and would "explore solutions to this type of dispute," which is a notably softer posture than the one it took after The New York Times filed in December 2023.
Why the destruction remedy is the part to watch
Damages are money. Model destruction is different. If a court ever granted it, the defendant would have to remove the infringing work from a training corpus and produce a model that never saw it, which in practice means retraining, not editing. There is no known method for surgically removing one newspaper's archive from a set of finished model weights.
That matters to anyone whose product sits on top of a commercial API. You do not choose your provider's training data and you cannot audit it, but you inherit the consequences of it. The realistic near-term outcomes look like this:
Outcome | Odds of hitting your app | What you would actually feel |
|---|---|---|
Cash settlement plus a licensing deal | Most likely | Nothing, or a modest price rise passed down later |
Injunction limiting future scraping | Plausible | Slower knowledge cutoffs, more stale answers about recent news |
Ordered model destruction or retraining | Least likely, highest impact | A model version disappearing on a legal timetable, not a product one |
Nobody should plan a business around the third row. But the first two rows are ordinary business risk, and they are worth the same kind of contingency thinking you would give any other single-vendor dependency. If you have never worked through what happens when a model you depend on gets deprecated, that exercise now has a second reason to exist.
What this does not change
It does not change whether the code or copy your own AI tools produce belongs to you. That question runs on separate law and separate contracts, and it is worth reading up on who owns AI-generated code rather than inferring anything from a training-data suit.
It also does not change your own obligations. If your app sends customer data to a model provider, the relevant question is still the boring one in your vendor's terms, not the one in the headlines: does this tool train on your data, and can you turn that off. Similarly, if you fine-tune or self-host anything, the model licence you accepted governs your position, and no court ruling about OpenAI's corpus will rewrite it for you.
The pattern behind the filing
This is the fourth broad category of AI copyright action to reach US courts: news publishers, book authors, image libraries, and music rights holders, each with slightly different theories. The Seattle Times and Newsday filing is notable mostly for stacking two mid-sized regional publishers together rather than fielding a single national name, which lowers the cost of participation for the next paper that wants in.
If you follow this space for practical rather than legal reasons, the signal to track is not the filings. It is whether provider terms of service start adding indemnification language for output, because that is the point at which the litigation risk has been priced and moved onto a balance sheet you can read. Watching release notes and terms pages beats watching court dockets for almost everyone building on these APIs.
FAQ
Does this lawsuit affect my ChatGPT or Copilot subscription today?
No. Nothing in a filed complaint changes a live service. Any operational change would follow a ruling or a settlement, and both are years away in cases of this shape.
Could OpenAI really be forced to delete a model?
It is the remedy the papers asked for, not the remedy a court has granted. Courts have broad discretion, and the practical difficulty of unwinding a trained model is an argument the defendants will make loudly.
Is my app exposed if the provider I use loses a case like this?
Directly, no. Your exposure is second-order: pricing, availability, and how fresh the model's knowledge stays. Treat it as vendor concentration risk, not legal risk.
Are more publishers likely to sue?
The two-plaintiff structure here makes joining cheaper for smaller papers, and coverage in both TechCrunch and Engadget notes this is part of a widening pattern rather than an isolated action.
How did this land?
About the author
Senior Editor, AI & Product
Cecilia leads the Swarmz editorial desk. She has spent a decade turning complex AI and product topics into writing people actually finish, and she owns the blog's quality bar.
