Advance Local Media LLC v. Cohere Inc., No. 25-cv-1305 (S.D.N.Y. 2025)

In another case addressing the use of copyrighted works by artificial intelligence technology (“AI”), the Southern District of New York denied a partial motion to dismiss a direct copyright infringement claim related to summaries generated by AI. The court held that the summaries could constitute copyright infringement and thus were sufficient to create a factual issue.

Defendant Cohere, a company in the business of developing, operating, and licensing artificial intelligence models, offers a generative model called Command that collects large amounts of text, articles, and other materials from the internet to generate outputs and summaries of those works, allegedly including copyrighted works.

On February 13, 2025, numerous media publishers, such as Condé Nast, The Atlantic, Forbes, The Guardian, the Los Angeles Times, Politico, and Vox Media, sued Cohere alleging, among other claims, copyright infringement for Command’s use of their copyrighted works. The publishers asserted that Command, when responding to user prompts, frequently reproduced their articles in full or in substantial part even when the material was behind paywalls. The model also generated outputs that reflected the sources verbatim and other times as “summaries” that closely followed the language and structure of the originals. The publishers provided seventy-five examples of such outputs, asserting that Command’s summaries often went beyond factual recitation to mirror the expression of their works. The publishers argued that these summaries served as “substitutes” for the original articles, harming the market for licensed content and readership revenue.

Cohere moved to dismiss, among other claims, the direct copyright infringement claim only to the extent Cohere was directly liable for generating “substitutive summaries.” Cohere argued that the summaries do not copy any protectable expression because it “incorporate[d] the abstracted facts into new and original sentences” and merely restated facts in new words. Even where the summaries do copy the publishers’ expressions, Cohere asserted that it did so minimally such that it was non-infringing.

The court rejected Cohere’s arguments and denied Cohere’s motion to dismiss on this basis. It held that the publishers plausibly alleged substantial similarity between their works and Command’s outputs, some of which allegedly copied entire paragraphs verbatim. The court found that the plaintiffs had plausibly alleged that Command’s outputs, even in the form of summaries, were substantially similar to the original articles and therefore could constitute infringing reproductions. The court also held that the publishers’ examples plausibly demonstrated copying beyond unprotectable facts.

This ruling joins a growing line of decisions, such as in New York Times v. OpenAI and Andersen v. Stability AI, in which courts have refused to dismiss claims against AI developers accused of misusing copyrighted works. Further, this case demonstrates that AI summaries may constitute infringement if expressive elements of the copyrighted works rather than facts are substantially used.

Have questions or need guidance? Connect with the authors of this article: Michael Bregenzer and Victoria Hanson.