Analysis
The Seattle Times and Newsday sued OpenAI and Microsoft on September 5 in the US District Court for the Southern District of New York, alleging that both companies trained AI systems on their journalism without permission or payment. The publishers described generative AI as "a snake eating its own tail" that could "destroy the very organizations" producing the original reporting it depends on, according to TechCrunch.
The uncomfortable detail is the prior relationship. Microsoft and OpenAI had funded some of The Seattle Times' journalism projects. A Microsoft spokesperson said the company was surprised by the filing and open to discussing a resolution -- which is roughly what a defendant says when it would rather license than litigate.
Two Camps, and a Third That's Now Choosing
This is the same courthouse and the same defendants as The New York Times' December 2023 suit, the case that opened this front. Since then the industry has split into two camps. One licenses: News Corp, Axel Springer, the Financial Times, the Associated Press and Reddit have all signed paid deals with OpenAI. The other sues: the Times, the Center for Investigative Reporting, The Intercept, Ziff Davis and a long list of authors and music publishers. A third group -- regional dailies owned by hedge funds and family trusts -- has mostly done neither, because litigation is expensive and their leverage is thin.
The Seattle Times and Newsday belong to that third group and are now choosing the courtroom. That matters more than the size of either paper. Regional publishers hold local archives that are hard to substitute and, in aggregate, represent a large share of the training corpus for anything resembling US local knowledge. If mid-market papers can sustain claims, the licensing price for the whole category resets upward.
The Counterweight
None of these cases has produced a definitive merits ruling on fair use for model training, and the ones that have moved furthest have narrowed rather than expanded. Judges have shown more sympathy for claims about pirated source corpora and verbatim regurgitation than for the broad proposition that training itself is infringement. A filing is not a judgment, and defendants with OpenAI's and Microsoft's resources can litigate for years while continuing to ship products.
For founders building on frontier models, the practical exposure is indemnification. Microsoft, OpenAI, Google and Anthropic all offer some form of customer copyright commitment, but the scope varies and the carve-outs matter -- particularly for outputs you have modified or for retrieval systems that surface source text. Read the specific indemnity in your enterprise agreement rather than assuming coverage, and know which party holds the duty to defend if a publisher names your product rather than your model vendor.
Pulse covered OpenAI's agent disclosure problems earlier this week, and the two threads connect: both are arguments about what a lab owes the outside parties affected by its systems. The copyright suits will be settled in dollars. The behavioral questions will not.
One date to circle: any SDNY consolidation of the newer newspaper claims with the NYT docket, which would give this group of plaintiffs the discovery record the original case has spent nearly three years building.
