News publishers have celebrated a victory in the first stage of their copyright lawsuit against Canadian AI start-up Cohere.
A judge has rejected in full Cohere’s motion to dismiss, saying publishers had “adequately alleged” that outputs from the AI provider were “quantitatively and qualitatively similar” to their content.
Some 14 news and magazine publishers are involved in the case, which was filed in February. They are: Advance Local Media, Conde Nast, The Atlantic, Forbes, The Guardian, Business Insider, LA Times, McClatchy Media Company, Newsday, Plain Dealer Publishing Company, Politico, The Republican Company, Toronto Star Newspapers, and Vox Media.
The publishers are all members of US trade association the News/Media Alliance.
Cohere produces large language model Command and other AI products marketed towards businesses.
The publishers have accused it of using their work to train its LLMs by copying and downloading text directly from websites and onto its servers via web crawlers and other bots.
Cohere attempted to dismiss most of the complaint against it, including part of the direct copyright infringement claim as well as the secondary infringement claim, but failed on all counts.
Judge Colleen McMahon, of the Southern District of New York, said in her ruling that Cohere is entitled to republish the “underlying facts” in the publisher content but that publishers provided enough examples showing Command’s outputs had copied and pasted their work that it creates a factual issue for a jury to consider.
The publishers provided 75 examples of Cohere’s alleged copyright infringement of which 50 allegedly included verbatim copies of their work and a further 25 examples of a mix of verbatim copying and “close paraphrasing”.
The ruling said the examples “reveal that, at least in some instances, Command delivers an output that is nearly identical to publishers’ works.
“For example, in response to the prompt ‘Tell me about the unknowability of the undecided voter,’ Command allegedly delivered an output which directly copied eight of ten paragraphs from a New Yorker article with very minor alterations.”
Cohere had argued that even where the AI summaries did copy some of publishers’ expression, they did so only minimally. The judge said this was an argument that should be heard at a later stage.
The ruling also found that publishers had sufficiently alleged that Cohere “had actual or constructive knowledge” that its products were infringing their copyright.
The publishers told the court they had put Cohere on notice that it was not authorised to use their works by including copyright notices and terms of service on their websites, as well as by sending do-not-crawl instructions to Cohere’s bots via robots.txt protocols.
They also claimed Cohere continued to unlawfully copy their work after they sent a cease-and-desist letter informing it of the alleged infringement.
Another way the publishers claimed Cohere is secondarily liable for copyright infringement is by inducement, meaning it actively encouraged direct infringement.
The publishers “plausibly pleaded”, the judge said, that Cohere did so by advertising Command as a tool to access news. A September 2023 marketing pitch said Command offered a “web search to access the latest news about trends and competitors”.
Announcing the Cohere AI app in September 2024, Cohere said it could “keep you up to date with the latest news”.
And a free online demo of Command “prepopulates the interface with a request to summarise recent
technology news, inviting prospective customers to use the models to access news stories”, the publishers said.
‘Hallucinations mislead users into thinking they are written by publishers’
The publishers have also alleged that Command sometimes delivers hallucinated text in response to a request for a specific article, with the output bearing trademarks of the publishers.
They said this “leads users to incorrectly believe that Command’s hallucinated articles are written by, associated with, or approved by publishers”.
Dismissing Cohere’s opposition to this part of the case, the judge said the publishers adequately alleged a likelihood of confusion on the part of users who would see their brands as “high-quality sources of reliable and informative content”.
The judge said: “According to publishers, it is not readily apparent to users that the output Command delivers is inaccurate, and users will therefore rely on the output as if it were publishers’ authentic content.
“This is especially likely given that publishers have publicly announced the licensing of their content to other AI companies in recent years. Based on publishers’ allegations, it is plausible that an ordinary consumer would be confused about an article’s source.”
‘First step towards justice for publishers’
Danielle Coffey, president and chief executive of the News/Media Alliance, said: “We’re grateful that Judge McMahon soundly rejected all aspects of Cohere’s motion to dismiss.
“As the complaint alleges, Cohere has engaged in massive, systemic infringement, and the affected publishers deserve their day in court.
“This decision is the first step towards justice for these publishers, who deserve the full legal protection offered by the law for their intellectual property.
“Copyright and trademark protections are vitally important for the health and longevity of the news media industry, which serves an important role in keeping society informed and supporting the free flow of information and ideas.”
A Guardian spokesperson told Press Gazette: “This is a confirmation of the strength of the claims we brought and the serious deficiencies in Cohere’s attempt to deflect the harm they are causing publishers.
“It is also further justification for the time and cost we have expended in bringing the suit and a sign that those efforts are worthwhile.”
The post News publishers win first round of copyright claim against AI start-up Cohere appeared first on Press Gazette.
