A federal judge in the Southern District of New York has allowed a coalition of major news publishers to move forward with claims that AI-generated news summaries may infringe copyright, marking a significant early ruling in litigation over generative AI outputs. In Advance Local Media LLC v. Cohere Inc., publishers including Forbes, Condé Nast, the Los Angeles Times, and The Atlantic allege that Cohere’s Command language models produce “substitutive summaries” that mirror the structure, narrative flow, and expressive choices of their original reporting. According to the complaint, some outputs closely track the publishers’ articles, occasionally near-verbatim, and in certain instances appear to circumvent paywalls, thereby competing directly with the underlying works.

Judge Colleen McMahon denied Cohere’s motion to dismiss, holding that the publishers plausibly alleged both direct and secondary copyright infringement, as well as related unfair-competition claims. The court rejected the notion that non-verbatim summaries are automatically lawful, emphasizing that infringement turns not on word-for-word copying but on whether an AI-generated summary appropriates the “qualitative” expression of the original. As the opinion notes, infringement cannot be measured solely by word count; courts must consider the overall expressive similarities, including narrative emphasis, structure, and stylistic elements.

The ruling does not resolve whether Cohere ultimately infringed; those questions will proceed to discovery, but it represents an important development for both AI companies and news organizations. For publishers, the decision opens a path to challenge AI outputs that may erode subscription or licensing revenue. For AI developers, it signals that even condensed or paraphrased outputs may carry copyright risk when they function as market substitutes for the original journalism.