Michelle
Morgante

By Michelle Morgante
Regional Managing Editor, The Recorder

  What You Need to Know

• The plaintiffs are members of the News/Media Alliance, a D.C.-based trade association.
• Cohere, based in Canada and San Francisco, has significant funding from Oracle, NVIDIA and Salesforce.
• The suit claims Cohere produces near-identical stories in real time, depriving the plaintiffs of subscriber and advertiser revenue.

Several of the country’s largest news publishers have filed suit against Cohere, an AI company that allegedly used their content to create near-verbatim copies in real time and also hallucinated error-filled stories that were falsely attributed to the plaintiffs.

The copyright and trademark complaint was filed Wednesday in the U.S. District Court for the Southern District of New York by Oppenheim and Zebrak of New York and Washington.

The plaintiffs are led by New York-based Advance Local Media, which owns daily newsrooms in several states, and Advance Magazine Publishers, which includes the Condé Nast publications Vogue, The New Yorker, Vanity Fair, Wired and others. Other plaintiffs are The Atlantic Monthly Group, Forbes Media, Guardian News & Media, Insider, Los Angeles Times Communications, The McClatchy Company, Newsday, Plain Dealer Publishing, Politico, The Republican Co., Toronto Star Newspapers and Vox Media.

The suit alleges Cohere routinely scraped content from the plaintiffs’ websites in order to train large language models for its own AI products. By doing so, Cohere competes with the publishers’ own ability to license their content with other AI companies and deprive it of revenue from subscribers and advertisers, the suit said.

“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,” it said.

Founded in 2019, Cohere is based in Toronto, Canada, and San Francisco. It was valued at $5.5 billion last year, with lead funders that include Oracle, NVIDIA and Salesforce, the suit said. Counsel for the defense has yet to appear.

In a statement, Cohere said it “strongly stands by its practices for responsibly training its enterprise AI.”

“We would have welcomed a conversation about their specific concerns—and the opportunity to explain our enterprise-focused approach—rather than learning about them in a filing. We believe this lawsuit is misguided and frivolous, and expect this matter to be resolved in our favor.”

The suit is the latest among several filed against AI companies by news organizations including The New York Times, Dow Jones and Thomson Reuters.

“This is a big case … perhaps bigger than the others that have been filed against AI companies,” said James Wheaton, who teaches journalism law at the journalism graduate schools of Stanford and the University of California, Berkeley.

What sets this complaint apart is the number of publications involved, as well as its new combination of copyright and trademark violations, Wheaton said in a phone interview with The Recorder.

“They show instances in which someone can go to the AI by Cohere, put in a search query that specifically references an article and get the article for free right in real time, meaning that this could easily displace subscriptions,” said Wheaton, who is not connected to the case.

“Why would anyone subscribe to a magazine or a newspaper if you could go that same day it's published and just download it for free or for some small fee?”

The suit points to Cohere’s “retrieval-augmented generation” feature, or RAG, which enhances the AI-produced content from the data used to train its LLM by supplementing it with articles copied in real time from other sources, such as from websites.

This reliance on trusted news sources has allowed Cohere CEO Aidan Gomez to boast that the “key differentiator” between its Command AI products and competitors is its ability to receive verifiable answers, the suit said.

“Indeed, Command’s RAG feature routinely returns copies of publishers’ copyrighted content as part of its responses to user queries, including same-day, breaking news pieces and verbatim copies,” it said.

“Left unfettered,” the suit said, “such misconduct threatens the continued availability of the valuable news, magazine, and media content that publishers produce. Indeed, Cohere admits that it seeks to supplant publishers, luring readers with a promise that its models can ‘keep you up to date with the latest news.’”

The trademark complaints allege that, when the RAG system is turned off, meaning it relies on previously ingested content, Cohere’s models invent false information, or hallucinations, that are then wrongly credited back to the publishers, harming their reputations for accurate, quality journalism and diminishing their brands.

One example cited in the suit said that, asked for information on the Oct. 7, 2023, massacre at the Nova Music Festival in Israel, Cohere produced information that confused details with a 2020 mass shooting in Nova Scotia, Canada, and created false content, including quotes attributed to a man who died in the 2020 attack, crediting the story to The Guardian.

“Cohere generates articles it attributes to the magazine or newspaper,” Wheaton said. “So, it's generating content and then putting the publication's name on it. It's hard to think of a more clear trademark infringement than that.”

In seeking a jury trial, the plaintiffs request an injunction to prevent Cohere from using the publishers’ copyrighted works and an order to have the company destroy all infringing copies of those works in its possession. It also seeks damages of up to $150,000 per work infringed.

Wheaton noted that, by filing the suit in New York, the center of the U.S. publishing industry, the plaintiffs have selected a district with experienced judges. “It’s looked upon generally as one of the best federal courts in terms of the judges who are there and the staffing it gets. And it’s the hometown for several of these companies.”

“I can’t speculate on what the motives of all these companies are,” he added. "I will say it was a very wise choice to choose the Southern District of New York.”

Lead counsel for the plaintiffs is Scott Zebrak. He referred a request for comment to the News/Media Alliance, a Washington-based trade association of which the plaintiffs are members.

News/Media Alliance President and CEO Danielle Coffey said that, as generative AI becomes more prevalent, legal protections are imperative to not only protect investments in the creative process but also to support the quality of what news consumers read and of the AI products themselves.

“As news, magazine and media publishers, we serve an important role in keeping society informed and supporting the free flow of information and ideas, but we cannot continue to do so if AI companies like Cohere are able to undercut our businesses while using our own content to compete with us,” Coffey said in a statement.

Roger Lynch, CEO of Condé Nast, said the company’s iconic brands “cannot live up to their exceptional standards if we allow their content to be stolen, distorted and trafficked. We will defend our rights fiercely and wherever they are infringed.”