Seattle Times and Newsday Sue OpenAI and Microsoft Over AI Training on Journalism

Two more major newspapers are taking OpenAI and Microsoft to court. The Seattle Times and Newsday filed a joint federal copyright infringement lawsuit, alleging that the companies scraped their journalism without permission to train AI models and that passages from reporting regularly appear in AI responses. The publishers are seeking destruction of any copies of their works, as well as the training datasets and AI models that incorporate them.

This is not an isolated action. The Seattle Times and Newsday join a list of nearly 400 local newspapers that recently sued OpenAI and Microsoft over similar claims, and the lawsuit echoes earlier cases filed by The New York Times, Ziff Davis, Merriam-Webster, and Encyclopedia Britannica. The central allegation is the same: AI training data sourced from copyrighted journalism without compensation or permission, and the resulting chatbots reducing the need for readers to visit publisher sites, costing subscription revenue.

What the lawsuit claims and seeks

The complaint, filed in the US District Court for the Southern District of New York, accuses OpenAI and Microsoft of copying the newspapers' journalism, including content behind paywalls, to train and operate AI products. Microsoft is named because Copilot is built on OpenAI's technology. The plaintiffs argue that AI systems reproduce passages from their reporting in responses to user queries, which directly undercuts the value of their work.

The requested remedy is aggressive. The publishers want an order requiring destruction of all copies of their works held by the companies, as well as the training datasets and AI models that incorporate them. If granted, that would mean unpublishing or retraining models to remove the copyrighted content, a far more invasive remedy than monetary damages alone.

Why AI builders should pay attention

This case is part of a broader legal shift that directly affects how AI companies source training data. If publishers succeed in proving that scraping paywalled journalism without permission constitutes infringement, the standard for training data collection could tighten significantly. For builders relying on open web scraping, the ruling could establish clearer boundaries around what is fair use and what requires licensing.

The lawsuit also puts a spotlight on Microsoft Copilot. Naming Copilot as a defendant product means the court may examine how downstream applications inherit liability from upstream models. Builders using foundation models from OpenAI or other providers should watch how courts treat the relationship between model provider and application layer.

The practical risks if publishers prevail

The destruction remedy is the most consequential part of the suit. It goes beyond asking for money; it asks a court to order the deletion of trained models. That would set a precedent that copyrighted training data can force model retraction or retraining, a costly and operationally disruptive outcome for any AI company. Even if the case settles before trial, the threat of such remedies may push AI companies toward more aggressive content licensing agreements.

For publishers, the suit argues that AI chatbots reduce site visits and article views, directly harming subscription revenue. This framing could gain traction if plaintiffs can show concrete traffic declines tied to AI-generated summaries.

Caveats and what remains unclear

The case is in its early stages. OpenAI and Microsoft have not yet responded to the complaint, and no court has ruled on the merits. The outcome will depend on how the court interprets fair use in the context of AI training, a question that remains unsettled across multiple pending cases.

The destruction remedy is an extreme ask and courts may be reluctant to order it, especially for models already deployed at scale. Even if the court finds infringement, the remedy could be limited to monetary damages or licensing requirements.

Additionally, the factual record is still developing. The specific evidence of reproduction and scraping will be tested in discovery. For now, builders should treat this as a signal that training data provenance is becoming a material legal risk, not just a compliance checkbox.