Toronto-based enterprise AI giant Cohere has filed a motion in a United States (US) Southern District of New York court asking it to dismiss a lawsuit in which media companies, including the Toronto Star, allege that the startup conducted large-scale copyright infringement. According to the lawsuit brought by the News/Media Alliance, Cohere scraped media content without consent and used it to train AI models, used media content in real time without authorization, and generated infringing outputs.
“The complaint offers a deafening nothing about real-world users, real-world prompts, or real-world outputs.”
Cohere motion
In its filing, Cohere claimed the media publishers had gone out of their way to “manufacture” a situation in which the company’s technology would generate allegedly infringing content. It disputed the notion that any practical infringement had taken place.
“The complaint offers a deafening nothing about real-world users, real-world prompts, or real-world outputs,” Cohere said in the motion.
BetaKit has asked Cohere and the News/Media Alliance for comment. In a blog post timed alongside the filing, Cohere maintained that the lawsuit showed a “lack of understanding” about the company’s enterprise product offerings, and that the plaintiffs used a developer demo tool in a way that bore “no semblance” to its intended uses.
Cohere further asserted in the post that its technology and customer base “respect intellectual property laws.”
The lawsuit, filed in February, comes from a mix of publishers that includes magazine conglomerate Condé Nast, Forbes, the Los Angeles Times, the Toronto Star, and Vox Media. The group seeks damages from Cohere that include as much as $150,000 USD ($207,000 CAD) for every infringement, as well as a court order barring Cohere from training its AI on these works.
RELATED: Major Canadian news orgs sue OpenAI for copyright infringement
Some publishers, including Condé Nast and Vox, already have licensing deals allowing ChatGPT creator OpenAI and other Cohere competitors to use their content. In January, Anthropic reached a deal with music publishers over its use of song lyrics. Other intellectual property (IP) holders are still embroiled in legal battles. OpenAI still faces a lawsuit from Canadian news outlets over alleged copyright abuses during the training of the model underlying the company’s flagship text generator ChatGPT.
Cohere’s chances of success with this motion, and the case in general, aren’t clear, Gilbert’s LLP Partner and IP litigation expert Paul Banwatt told BetaKit in an interview. While he said copyright holders have been “fairly successful” so far, he noted that there isn’t a significant legal framework in place for AI-related IP issues, and that it was likely this case will be “precedent setting.”
“As a general matter, what’s probably needed here is law and regulation,” Banwatt said.
He added that copyright law in the US has an “absolute nature” regarding violations. Unless the alleged infringer can find an exception, it doesn’t matter whether or not that party intended to infringe copyright.
“Those sorts of things stop mattering,” Banwatt said.
Cohere launched in 2019 and focuses on development of large-language model (LLM) products aimed at businesses, where OpenAI, Anthropic, Google, and other rivals target general consumers. It launched an Embed 4 model in mid-April that helps AI agents search for and retrieve data, and earlier this month forged partnerships with Dell and SAP.
The bid to dismiss the lawsuit comes soon after Cohere reported doubling its annualized revenue to over $100 million USD ($138 million CAD). In a Bloomberg interview, CEO Aidan Gomez maintained that his business was “very, very close” to profitability. However, The Information reported that Cohere had still achieved only 15 percent of its revenue goal projections from 2023.
Feature image of Aidan Gomez, CEO of Cohere, courtesy of World Economic Forum on Flickr (CC BY-NC-SA 2.0).