Canadian company Cohere, star of AI for businesses, sued for infringement by press in New York courts

On 13 February 2025, a consortium of 14 leading international publishers, including Condé Nast, Forbes, The Atlantic and The Guardian, filed a lawsuit against Cohere Inc. alleging numerous copyright and trademark infringements. The suit was filed in the U.S. District Court for the Southern District of New York, which is already seized of the well-known ongoing case between OpenAI and The New York Times. The publishers accuse Cohere of using more than 4,000 copyrighted works without permission to train its AI models, and of displaying news articles without directing users to the original publishers' websites. 

Cohere is a Toronto-based Canadian unicorn specialising in AI for compagnies. It was founded in 2019 by, three former Google Brain employees, Aidan Gomez, Ivan Zhang and Nick Frosst. It allows companies to integrate its AI platform on public clouds (e.g. Google Cloud, Amazon Web Services), virtual private clouds or even locally. In July 2024, the company raised $500 million, with a valuation of $5.5 billion. Cohere now competes with other AI targeting businesses, such as Microsoft's Copilot and Google's Vertex AI

A group of publishers has formed a consortium to strengthen its case against Cohere. The consortium includes a number of general newspapers and magazines from the US (The New Yorker, Vogue, Forbes, Politico), Canada (Toronto Star Newspapers) and the UK (The Guardian).

The plaintiffs claim that Cohere's practices not only infringe their intellectual property rights, but also undermine their business models, which rely heavily on advertising and subscription revenues. They argue that by providing users with content that closely mirrors or reproduces their original articles, Cohere diverts traffic away from their platforms, impacting their financial viability. In addition, the complaint highlights instances where Cohere's AI has generated "hallucinated" content - articles fabricated and attributed to publishers - that could damage their reputations.

In response, Cohere refuted the allegations, calling the lawsuit "misguided and frivolous". A company spokesperson said Cohere maintained its practices of responsibly training its corporate AI and had controls in place to mitigate the risk of intellectual property infringement. The company expressed a preference for direct dialogue to address publishers' concerns rather than litigation

The main legal grounds for the publishers' complaint include:

1. Copyright Infringement (17 U.S.C. § 101, et seq.) - Cohere allegedly copied and used the publishers' copyrighted works in its training and generation of AI results without a license from the publishers.
2. Trademark infringement (15 U.S.C. § 1051, et seq. - Lanham Act) - Cohere is accused of falsely attributing AI-generated content to publishers, thereby misleading the public.
3. Unfair competition - The publishers claim that Cohere is engaging in unfair competition by using protected journalistic content to provide similar services without compensating the original creators.

‍

1. Infringement of publishers' copyright

The publishers accuse Cohere of copyright infringement, accusing it of copying and exploiting their content without authorisation. According to the complaint, Cohere recovered and duplicated more than 4,000 articles, including paid content, directly from the publishers' websites. These works would then have been used to train its artificial intelligence models, the results of which would have included text extracts or summaries of the original articles, thereby replacing the legitimate sources.

The evidence produced by the publishers consists of screenshots of Cohere's AI, showing in particular that when users request a summary of an article, the article is not only summarised but is also available in full in the AI's "Under the Hood" section. The 'Under the Hood' section displays the AI's reasoning in uncoded language :

‍

‍

In some cases, Cohere's AI seems to 'regurgitate' newspaper articles, modifying them to a greater or lesser extent. In some cases, the abstracts are copied almost word for word from the editors' articles:

‍

‍

As a result, the publishers are claiming compensation for the damage suffered on the licensing market. They argue that Cohere's practices undermine established and emerging business models for the licensing of news content. While other players in the sector, such as OpenAI, have negotiated licensing agreements with various publishers, Cohere has not taken any steps in this direction. 

The publishers are also denouncing Cohere's misuse of retrieval-based augmented generation (RAG), which they see as a form of ongoing counterfeiting. This process enables Cohere's AI to retrieve data in real time from the Internet in order to enrich its responses. However, the complaint claims that this technology is used to extract and actively reproduce articles from publishers' sites, which would constitute repeated and systematic copyright infringement.

‍

2. Infringement of publishers' trademarks 

The publishers are also accusing Cohere of infringing their trademarks, which correspond to press titles. The complaint alleges that Cohere's AI generates false articles that it wrongly attributes to legitimate media, thereby misleading the public.

For example, in the "Under the Hood" section, the sources of the articles are indicated for each content item using the publishers' marks on the content generated by the AI.

‍

‍

According to the plaintiffs, this artificial content gives the illusion that it emanates from credible sources such as The Guardian, Forbes or Politico. In addition, the unauthorised use of the publishers' brands and titles (Vogue, The New Yorker, Wired, Forbes, The Atlantic, Politico) to identify this generated content would constitute both brand dilution and false association, damaging the reputation and editorial integrity of the media concerned.

‍

3. Damages based on competition law and unjust enrichment

According to the plaintiffs, Cohere competes directly with publishers by using its AI to generate summaries of news articles that discourage users from consulting the original sites.

The complaint alleges that Cohere exploits copyrighted content without paying compensation to the creators, thereby constituting unjust enrichment. By providing easy access to information without redirecting to official sources, the results produced by Cohere's AI would reduce users' incentive to subscribe to publishers, thereby undermining publishers' revenues.

Publishers are asking :

1. An injunction to prevent Cohere from continuing to use their copyrighted works.
2. Damages for copyright and trademark infringement in the amount of $150,000 per infringing article or work (i.e. at least $600 million) or alternatively the profits that Cohere would have made from the infringement (to be determined at trial). 
3. Destruction of articles and results based on unauthorised data.

‍

4. Case law on AI

This case is reminiscent of the New York Times v. OpenAI lawsuit before the same United States District Court Southern District Of New York. The New York Times v OpenAI lawsuit, filed in December 2023, is based on accusations that OpenAI and Microsoft used millions of copyrighted New York Times articles to train their artificial intelligence models, including ChatGPT, without permission. In January 2025, OpenAI and Microsoft filed a "motion to dismiss", arguing that their use of Times content falls within the scope of fair use. At this stage of the proceedings, the arguments have been heard by a federal judge, but no decision has yet been handed down. In this context, it is clear that AI companies will rely on the fair use doctrine to justify training their models on protected content. It is therefore foreseeable that Cohere will adopt a similar defence in the action brought against it

If the publishers win their case, AI companies may have to seek licensing agreements before training models on copyrighted works.

‍

To be continued...