Key Points  

• On February 13, 2025, Cohere Inc (Cohere) was sued by a number of news publishers for copyright and trademark infringement 

  
  

• The Cohere case is very similar to the Thomson Reuters case, where Thomson Reuters was successful and granted partial summary judgment for the infringements and fair use claims 

  
  

• The Thomson Reuters case is instructive when it comes to the Cohere and the New York Times cases   

On February 13, 2025, Cohere Inc (Cohere) was sued by a number of news publishers, including Advance Local Media LLC; Advance Magazine Publishers Inc. D/B/A Conde Nast; The Atlantic Monthly Group LLC; Forbes Media LLC; Guardian News & Media Limited; Insider, Inc.; Los Angeles Times Communications LLC; The McClatchy Company, LLC; Newsday LLC; Plain Dealer Publishing Co.; Politico LLC; The Republican Company; Toronto Star Newspapers Limited; and Vox Media, LLC (Publishers). 

What is the lawsuit about? 

The first paragraph of the lawsuit discusses the nature of the case, and points out that the lawsuit is about protecting journalism from systemic copyright and trademark infringement:  

“Rather than create its own content, Cohere takes the creative output of Publishers, some of the largest, most enduring, and most important news, magazine, and digital publishers in the United States and around the world. Without permission or compensation, Cohere uses scraped copies of our articles, through training, real-time use, and in outputs, to power its artificial intelligence (AI) service, which in turn competes with Publisher offerings and the emerging market for AI licensing. Not content with just stealing our works, Cohere also blatantly manufactures fake pieces and attributes them to us, misleading the public and tarnishing our brands” 

In fact, the lawsuit clearly talks about how publishers spend enormous amounts of time investigating, reporting, and ultimately publishing their expressive and groundbreaking pieces, which span the full spectrum of investigative reporting, breaking news, opinion pieces, arts and entertainment reviews, sports coverage, and political and business journalism.  

The Publishers claim that Cohere, with its valuation of over $5 billion, fails to license the content it uses and takes the Publishers’ valuable articles, without authorization and without providing compensation. It copies, uses, and disseminates the Publishers’ news and magazine articles to build and deliver a commercial service that mimics, undercuts, and competes with lawful sources for their articles and that displaces existing and emerging licensing markets. 

More specifically, the Publishers claim that Cohere copies the Publishers’ works to train its suite of LLM AI systems of products. They claim that the Publishers value innovation and AI if ethically deployed—they already license their articles to AI companies. However, they say that Cohere improperly usurps their creative labour and investments for the sake of its own profits.  

Most troubling, the Publishers claim that Cohere’s AI models deliver outputs that include full verbatim copies, substantial excerpts, and substitutive summaries of Publishers’ works—even current, breaking news pieces and articles protected by paywalls.  

Ultimately, the Publishers claim that Cohere’s actions amount to “massive, systematic copyright infringement and trademark infringement, and have caused significant injury to Publishers”. Moreover, they are adamant that left unfettered, such misconduct threatens the continued availability of the valuable news, magazine, and media content that Publishers produce. In fact, there are over 4,000 articles that the Publishers claim are registered copyrighted works that have been infringed. Worse, they claim that Cohere has passed off its own hallucinated articles as articles from the Publishers. 

The following is the list of the Publishers’ Causes of Action: 

• Count I: Direct Copyright Infringement, in violation of the Copyright Act—each infringement constitutes a separate and distinct act of infringement, and Cohere’s acts of infringement are willful, intentional, and purposeful, in disregard of and with indifference to the Publishers’ rights 

• Count II: Secondary Copyright Infringement, in violation of the Copyright Act—to the extent Cohere seeks to shirk responsibility for its own conduct by shifting blame onto its users and customers, the Publishers also bring claims for secondary liability in the alternative 

• Count III: Trademark Infringement, in violation of the Lanham Act—Cohere uses marks that are either identical to, variations on, or colourable imitations of Publishers’ federally registered trademarks in connection with the generation and distribution of hallucinated articles that Publishers did not publish. Cohere has caused and is likely to cause confusion, mistake, or deception as to whether the hallucinated articles Cohere provides are associated or affiliated with, or are sponsored, endorsed, or approved by the Publishers 

• Count IV: False Designation of Origin, in violation of the Lanham Act—Cohere has used and continues to use the Publishers’ marks in interstate commerce in a misleading manner, falsely associating the Publishers’ valuable trademarks and trusted brands with Cohere and Cohere’s products and services. As a result, Cohere’s users are deceived and are likely to continue to be deceived by the appearance of the Publishers’ trademarks on Cohere’s hallucinated articles 

  
  

The Publishers are asking for judgment that Cohere is liable under the Copyright Act and the Lanham Act; equitable relief including a permanent injunction; an order telling Cohere to stop training or fine-tuning AI models or generating content from AI models; an order requiring Cohere to destroy under the court’s supervision all infringing copies of the Publishers’ works; statutory damages and actual damages; fees; and interest. 

Who is Cohere, and What is its Reaction to the Lawsuit? 

Cohere is a Canadian company with its principal places of business in Toronto, San Francisco, London (UK), and New York. It is a multinational technology company focused on AI for the enterprise, specializing in large language models. 

It has been reported that Cohere’s response is that the company expects that the court will side with Cohere because it has long worked to mitigate the risk of intellectual property infringement. Further, Josh Gartner apparently said that Cohere “strongly stands by its practices for responsibly training its enterprise AI” and believes the lawsuit is “misguided and frivolous.” 

This case is part of a string of intellectual property cases against AI companies 

As you may recall, I wrote an article titled, New York Times Sues OpenAI and Microsoft for Copyright Infringement, which discussed the copyright lawsuit that the New York Times launched against OpenAI and Microsoft for the same type of alleged infringement.  

This case has not yet been decided, and as we shall see below, the New York Times could very well win its case. 

A Similar Case that Supported a Copyright Holder  

In another similar case, Thomson Reuters and West Publishing Group sued Ross Intelligence in a District Court for the District of Delaware. 

This February 11, 2025 decision is interesting, and uniquely written by Bibas, the Circuit Judge. This was his first paragraph in the decision: 

“A smart man knows when he is right; a wise man knows when he is wrong. Wisdom does not always find me, so I try to embrace it when it does––even if it comes late, as it did here” 

This was because he actually revised his previous 2023 decision and ultimately granted Thomson Reuter’s motion for partial summary judgment on direct copyright infringement and related defenses, and on fair use. To that end, he denied Ross’s motion for summary judgment on fair use and Ross’s motion for summary judgment on Thomson Reuters’s copyright claims. 

What happened in this case? As we know, Thomson Reuters owns one of the largest legal research platforms, Westlaw. Users need to pay to access and use the platform. Westlaw also contains editorial content and annotations such as the headnotes that summarize key points of law and case holdings. Westlaw organizes its content using the Key Number System, a numerical taxonomy—Thomson Reuters owns copyrights in Westlaw’s copyrightable material.  

Ross decided to make a legal research search engine that used AI and that competed with Westlaw. Ross needed a database of legal questions and answers to train the tool; therefore, Ross asked to license Westlaw’s content. Thomson Reuters refused.  

Consequently, Ross made a deal with LegalEase to get training data in the form of “Bulk Memos”, which are lawyers’ compilations of legal questions with good and bad answers. Notably,  LegalEase gave those lawyers a guide explaining how to create those questions using Westlaw headnotes, while clarifying that the lawyers should not just copy and paste headnotes directly into the questions. LegalEase sold Ross roughly 25,000 Bulk Memos, which Ross used to train its AI search tool.  

In response, Thomson Reuters sued for copyright infringement. More specifically, the company sued for direct copyright infringement and claimed that Ross’s defense of fair use was unsuccessful.  

When it came to direct copyright infringement, the judge stated that Thomson Reuters had to show both that (1) it owned a valid copyright and (2) Ross copied protectable elements of the copyrighted work. The second element required showing that Ross actually copied the work and that its copy was substantially similar to the work.  

The judge granted summary judgment for Thomson Reuters on whether the headnotes and the Key Number System were original enough to prevent Ross from rebutting any presumption of validity. Looking at about 4,000 headnotes, the judge talked to experts and compared judicial decisions to the headnotes, and decided that Thomson Reuters should be granted summary judgment on actual copying of the data. 

Then, the judge asked whether an ordinary user of a product would find it substantially similar to the copyrighted work, and answered in the affirmative: Thomson Reuters was granted summary judgment on substantial similarity regarding the headnotes. 

Moreover, the judge confirmed that Ross’s defenses failed—all of them. He swiftly rejected the innocent infringement claim since innocence did not limit liability. Similarly, he disagreed with Ross that Thomson Reuters misused its own copyrights. He also rejected Ross’s claim involving a merger of expression. He also quickly rejected a claim of scenes à faire (a principle in copyright law in which certain elements of a creative work are held to be not protected when they are mandated by or customary to the genre). 

But the interesting defense that was raised by Ross was the fair use defense. Section 107(1–4) of the Copyright Act required the judge to consider at least these four factors in the analysis: 

1. The use’s purpose and character, including whether it is commercial or nonprofit: this one went to Thomson Reuters since Ross’s use was commercial and not transformative 

2. The copyrighted work’s nature: this one went to Ross since Westlaw’s work was not that creative 

3. How much of the work was used and how substantial a part it was relative to the copyrighted work’s whole: this one went to Ross since the judge stated that “What matters is not “the amount and substantiality of the portion used in making a copy, but rather the amount and substantiality of what is thereby made accessible to a public for which it may serve as a competing substitute”  

4. How Ross’s use affected the copyrighted work’s value or potential market (this was the most important factor): this one went to Thomson Reuters since Ross tried to compete with Westlaw by developing a market substitute, and it did not matter whether Thomson Reuters had used the data to train its own legal search tools—the effect on a potential market for AI training data was enough 

   
   

Thus, when balancing the factors, it became clear that Thomson Reuters had to be successful. The judge granted partial summary judgment to Thomson Reuters on direct copyright infringement for the headnotes. For those headnotes, the only remaining factual issue on liability was that some of those copyrights may have expired or been untimely created. This factual question underlying copyright validity was saved for the jury. The judge also granted summary judgment to Thomson Reuters against Ross’s defenses of innocent infringement, copyright misuse, merger, scenes à faire, and fair use. Likewise, the judge denied Ross’s motions for summary judgment on direct copyright infringement and fair use. Though this newer decision replaced many parts of the 2023 decision, some parts remained relevant, including rulings on contributory liability, vicarious liability, and tortious interference with contract. 

What does this mean for AI and copyright? 

As can be seen from the Thomson Reuters case, the defense of fair use is not likely to be successful for defendants who directly copy material and use it to train AI models. We see that judges complete a side-by-side comparison of the products at issue to make their decisions. The defendants would have to show that their new work was transformative. This decision may be instructive for the upcoming decisions related to the New York Times and Cohere. 

In fact, judges who examine the following four key factors in the fair use analyses go through each fact situation to make a determination: 

1. The use’s purpose and character, including whether it is commercial or nonprofit 

2. The copyrighted work’s nature 

3. How much of the work was used and how substantial a part it was relative to the copyrighted work’s whole  

4. How the use affected the copyrighted work’s value or potential market (this is the most important factor) 

   
   

It will be interesting to see if the New York Times and the Publishers will be successful in light of this Thomson Reuters decision. We will keep you posted…