A group of fourteen major publishers, including The Guardian, Condé Nast, The Atlantic, Politico, LA Times, Business Insider, and Forbes, has filed a lawsuit against AI company Cohere for alleged copyright and trademark infringement. The lawsuit, filed in New York’s Southern District, claims the Canadian company used thousands of their articles without authorization to train and operate its AI systems.
Publishers Seek Injunction and Damages for Copyright Infringement
The publishers are seeking damages of up to $150,000 per infringed work and a permanent injunction. They argue that Cohere is using their content without permission or compensation to power AI services that compete with publishers’ offerings and emerging AI licensing markets.
Cohere Responds to Infringement Allegations
Cohere has responded by defending its practices. The AI company called the lawsuit “misguided and frivolous” and stated they have put controls in place to mitigate IP infringement risks. They said they regretted not having the opportunity to discuss the publishers’ concerns before legal action was taken.
Paywalled Content and Hallucinations
The lawsuit highlights specific examples of alleged infringement. This includes instances where Cohere’s AI provided full access to paywalled content and generated “hallucinations” — fabricated articles falsely attributed to the publishers. One notable example involved a confused and entirely fictional article about the Nova Music Festival massacre, which mixed details with an unrelated shooting in Nova Scotia.
Increasing AI Copyright Disputes
This legal action is part of a growing trend of copyright disputes in the AI industry. AI models require an enormous amount of content for training purposes. Similar lawsuits have been filed by The New York Times against OpenAI and by News Corp.’s Dow Jones against Perplexity AI. In a recent significant precedent, Thomson Reuters won the first major AI copyright case against Ross Intelligence. A judge ruled the startup had infringed copyright law by reproducing content from Reuters’ Westlaw database.