Summary: The Seattle Times and Newsday have sued OpenAI and Microsoft over alleged copyright infringement. The most unusual part is not the damages claim but the request to destroy copies, training datasets and even models to the extent that they contain the publishers' protected works.
For publishers, this is more than another copyright case. The complaint connects training, paywalls, reproduced content and potential loss of website traffic into one chain of harm. That connection may matter more to the future of AI search and news distribution than the narrow question of whether one training dataset was lawful.
Confirmed
The lawsuit was filed on September 4, 2026 in the U.S. District Court for the Southern District of New York. The case number is 1:2026cv07644. The plaintiffs are The Seattle Times Company and Newsday LLC; the defendants include multiple OpenAI entities and Microsoft.
According to Reuters, the newspapers allege that the companies copied journalism without permission for use in AI systems, including content behind paywalls. Products named in the allegations include ChatGPT, Microsoft Copilot and AI features in Bing.
The legal distinction matters: these are currently allegations by the plaintiffs, not findings by the court. OpenAI generally argues that its models are trained on publicly available data under fair-use principles. Microsoft said it was surprised by the lawsuit and indicated that it was open to discussions.
Why model destruction stands out
Copyright disputes are often discussed mainly as a financial question: how large was the loss, and should a licence have been paid? The requested remedy here goes further. The plaintiffs are also seeking destruction or impoundment of affected copies, training datasets and models.
That does not mean a court will grant such relief. Technically, it would also matter whether specific works can be tied to a particular model, training run or retrieval system. Still, the request raises the stakes because it targets not only past use but potentially systems that have already been built.
The mosaic: training, output and traffic become one problem
The case is notable because several disputes that are usually treated separately are combined. First, there is the question of whether content was copied for training or operation. Second, the publishers allege that AI systems can reproduce or closely paraphrase articles. Third, they argue that those answers can substitute for visits and subscriptions.
That third layer is strategically important for news companies. A publisher can be correctly cited by an AI system and still lose economically if the user receives a complete answer and never opens the original source. Conversely, AI systems can create new reach if the source contains clear additional value that a short generated answer cannot replace.
The argument therefore moves beyond training and into market structure: will AI become an information layer that replaces original sources, or a distribution layer that sends readers to them?
Pandorex Analysis
For small and mid-sized publishers, that second question is probably more important than the dramatic demand to destroy models. Even a complete legal victory against specific training practices would not automatically reverse the economics of AI search. Google, ChatGPT, Gemini and other systems will continue trying to assemble direct answers.
The more durable pressure point is likely to be attribution, licensing, opt-out rules, compensation and measurable referral traffic. Publishers that produce original research become more valuable as sources for AI systems. Publishers that mostly rewrite other outlets are easier to replace.
That is directly relevant to Pandorex: the strongest defence against becoming interchangeable is not avoiding AI in the writing process, but creating original information value through primary sources, patent analysis, technical comparisons and connections between multiple signals.
What argues against this interpretation
The lawsuit does not prove that OpenAI or Microsoft acted unlawfully. It is also unclear whether a court would consider broad destruction of models technically or legally proportionate. Fair-use questions in the United States remain highly fact-specific, and different AI cases may reach different outcomes.
Falling publisher traffic also cannot automatically be attributed to a single AI platform. Changes in Google Search, social media, user behaviour, paywalls and general news consumption all operate at the same time. The case is therefore an important signal, but not yet a verdict on the future relationship between AI and journalism.
Pandorex View: The strategic impact of this case is less about a realistic short-term “deletion of ChatGPT” and more about growing pressure on platforms to handle provenance, licensing and referral of news content in a technically cleaner way.