Background
The Seattle Times and Newsday filed federal lawsuits on Tuesday, accusing OpenAI and Microsoft of illegally harvesting their copyrighted articles to train large language models (LLMs) such as ChatGPT. The complaints mirror earlier actions by The New York Times, The Associated Press, and other outlets that claim their content was scraped without permission, violating the Copyright Act and the Digital Millennium Copyright Act.
Both plaintiffs allege that OpenAI’s data‑collection pipeline scraped millions of pay‑walled and subscription‑only stories, then fed them into the training corpus that powers ChatGPT‑4 and the upcoming GPT‑5. They argue that the companies profited from their journalism while offering no licensing fees, attribution, or opt‑out mechanism.
Legal claims
According to the filings, the defendants “systematically accessed and downloaded copyrighted works” in breach of Section 106 of the Copyright Act. The suits seek injunctive relief to halt further training on the plaintiffs’ content, monetary damages for past use, and a permanent injunction against future scraping.
OpenAI and Microsoft have responded with a brief motion to dismiss, asserting that the training data falls under “fair use” because it is transformed for a different purpose and that the models do not reproduce the original articles verbatim. The companies also point to the lack of a clear legal framework governing AI training data, a point that courts have yet to settle.
Why it matters to developers
These lawsuits signal a shift from abstract policy debates to concrete litigation that could reshape the data pipelines behind every commercial LLM. If courts side with the publishers, developers may be forced to obtain explicit licenses for any copyrighted text used in training, dramatically increasing costs and slowing iteration cycles.
Beyond cost, the legal risk extends to downstream products. A model trained on unlicensed data could be deemed infringing if it reproduces protected passages, exposing SaaS providers, API users, and even open‑source contributors to liability.
Actionable steps for developers and founders
- Audit data sources. Conduct a thorough inventory of the text, code, and image corpora used to train or fine‑tune models. Identify any content that is not clearly in the public domain or covered by an open license.
- Secure licenses or opt‑outs. Reach out to publishers, news aggregators, and rights holders to negotiate licensing agreements. Implement automated opt‑out mechanisms where possible.
- Document provenance. Keep detailed logs of when, where, and how each dataset was acquired. This documentation will be critical if you need to demonstrate good‑faith compliance.
- Consider synthetic or public‑domain data. Augment training sets with data generated by other models, public‑domain texts, or datasets released under permissive licenses such as CC‑0.
- Monitor legal developments. Follow the Seattle Times/Newsday case and related rulings closely. Adjust your compliance roadmap as courts issue precedents on fair use and AI training.
- Update terms of service. If you expose an API, clarify that users must not use the model to reproduce copyrighted text without permission.
Current litigation landscape
| Publisher | Defendant(s) | Key Claim |
|---|---|---|
| The Seattle Times | OpenAI, Microsoft | Unauthorized scraping for LLM training |
| Newsday | OpenAI, Microsoft | Same as above |
| The New York Times | OpenAI | Copyright infringement, unfair competition |
| The Associated Press | OpenAI | Massive unlicensed text use |
Outlook
The outcome of these cases will likely define the boundary between “fair use” and “copyright infringement” for AI training data. In the meantime, developers should treat the lawsuits as a warning sign: the era of freely scraping the web for training material is ending, and proactive compliance will become a competitive advantage.