Remember when the New York Times sued OpenAI and Microsoft, and a lot of people shrugged it off as one big paper flexing its legal muscles? That framing has aged badly. What looked like a one-off has turned into a pattern, and this week the pattern picked up two more names.
The Seattle Times and Newsday have filed suit against Microsoft and OpenAI, alleging the companies used their copyrighted journalism without permission to train AI models. They want damages, and they want their content pulled out of the training datasets entirely.
I write about AI agents for people who don’t build them, so let me explain why a copyright case involving two regional newspapers matters if you’ve ever asked a chatbot the news for you.
What “training on journalism” actually means
When people hear that an AI model was “trained on” news articles, they often picture something like a library, where the AI keeps copies on a shelf and looks things up when asked. That’s not quite it.
Training is closer to reading millions of examples and absorbing patterns from them. How sentences fit together. How a city council story is usually structured. What facts tend to cluster around a given topic. The model doesn’t store the articles as files, but it does come out the other side able to produce text that reflects what it read.
That distinction is the entire legal fight. AI companies have generally argued that learning patterns from text is different from copying it. Publishers argue that you can’t build a commercial product out of someone else’s work without paying for it, no matter what happens inside the black box.
Why the plaintiffs’ names matter here
The Seattle Times and Newsday aren’t national wire services. One covers the Pacific Northwest, the other covers Long Island. Local and regional papers operate on thinner margins than the big nationals, and their value comes from coverage nobody else produces: school board meetings, zoning fights, regional politics.
The lawsuit argues that OpenAI’s models are causing substantial harm to media businesses. If a reader can get the substance of a local story from an AI assistant without ever visiting the paper’s site, the paper loses the traffic, the subscription prospect, and the ad revenue that funded the reporting. The reporting still cost money to produce. The revenue just went somewhere else.
The agent angle nobody’s talking about enough
Here’s where this gets specific to the stuff I write about. AI agents are systems that don’t just answer questions but go out and do multi-step work on your behalf: search, read, compare, summarize, and report back.
An agent asked “what’s happening with the Seattle housing debate” doesn’t send you a list of links. It reads the coverage and hands you a synthesis. That’s genuinely useful. It’s also a fundamentally different relationship between reader and publisher than the one the web was built on.
Search engines sent you somewhere. Agents bring the answer to you. When the destination stops getting visitors, the economics of producing the thing at the destination stop working. These lawsuits are, in part, an argument about who pays for that shift.
What to actually watch for
You don’t need to follow court filings to track where this goes. A few things to keep an eye on:
- Licensing deals instead of verdicts. Several publishers have signed paid agreements with AI companies rather than fighting in court. Every new lawsuit strengthens the negotiating position of publishers who’d rather sign a deal.
- Whether “remove our content” is even possible. The Seattle Times and Newsday are asking for their work to be taken out of the training data. Removing specific material from an already-trained model isn’t a simple delete operation, and how courts handle that request could shape technical requirements for years.
- Citation behavior in the tools you use. Watch whether the assistants and agents you rely on start linking sources more prominently. That’s often a legal and business response, not just a design choice.
- Which publishers join next. Two more names this week. The list has been getting longer, not shorter.
My honest read
I don’t think this ends with AI companies losing the ability to train on text. I think it ends with a price attached to it, negotiated unevenly, with the biggest publishers getting the best terms and smaller outlets getting whatever’s left. That’s how these things usually settle.
But the questions being raised are the right ones, and they’re worth caring about even if you never read a court document. The tools many of us now use to understand the world were built on work that somebody had to fund, assign, report, and edit. If that funding model breaks, the tools get worse too, because there’s less real reporting flowing into them.
Two regional newspapers just made that argument in federal court. Whatever the outcome, the argument itself is one the rest of us should be paying attention to.
🕒 Published: