A group of news publishers, including the Toronto Star, are suing Toronto-based AI company Kohia for copyright infringement.Dado Luvitch/Reuters
A U.S. court has denied a motion by Toronto-based artificial intelligence company Kohia to dismiss copyright lawsuits brought against it by major news organizations, including the Toronto Star.
A lawsuit filed in New York accuses Cohere of violating copyright by illegally scraping news content to build its AI models. The company’s tools are said to be able to reproduce articles word-for-word and provide substantive summaries in some cases, even when the articles are behind a paywall.
In May, Cohere, which has a market capitalization of $7 billion, filed a motion asking the court to dismiss certain claims, alleging that the publisher had “deliberately misused” its tools to “manufacture” the lawsuit.
Judge Colleen McMahon rejected Kohia’s motion in a ruling issued Thursday. The ruling does not address the nature of the allegations, but rather whether the newspaper company properly filed the lawsuit.
Kohia declined to comment, but in a statement earlier this year called the lawsuit “misguided and frivolous.”
News publishers include Condé Nast, The Atlantic, Forbes Media, and McClatchy Media Company. The Toronto Star is the only Canadian plaintiff.
“As the complaint alleges, Cohere engaged in large-scale, systematic violations, and the affected publishers deserve their day in court. This decision is a step toward justice for these publishers,” News and Media Alliance President and CEO Daniel Coffey said in a statement Friday. The nonprofit organization helped organize the lawsuit on behalf of its members.
How AI is reshaping the art and science of cooking
The lawsuit is one of many brought by publishers, artists, and other creators against AI companies that use copyrighted material to build models without consent or payment. AI companies argue that such practices are allowed under copyright law through provisions known as fair dealing or fair use, but how it applies to building commercial AI models has proven controversial.
Some AI companies have payment agreements with news organizations and other companies to train their materials.
The lawsuit against Cohere alleges the use of copyrighted material in building AI models, but the company only asked the court to dismiss claims that its tools produce material that infringes copyright.
According to the ruling, news publishers identified 75 instances of alleged infringement based on Kohia’s model, including 75 instances that included verbatim text and 25 instances that included a mix of copying and near-paraphrasing. Kohia argued that the examples in the lawsuits differ in writing style, tone, length, and sentence structure.
The court disagreed, stating that in some cases the output was “substantially identical” to the original work. “The publishers allege that Kohia designed the system to do just that. These allegations are sufficient to raise questions of fact for the jury’s consideration,” Judge McMahon wrote.
Kohia also maintained that its AI tools were designed to improve productivity in companies and that no customers would use its systems to intentionally infringe copyrights, writing that the media outlets were “constructing a scenario that would not occur” in the real world.
Judge McMahon wrote that some of Kohia’s own marketing promotes its tools as a way to stay up to date on the latest news, and that the publisher “plausibly alleges that Kohia is taking active steps to facilitate infringement.”
Another of the publishers’ arguments is that Cohere’s model “hallucinates” misinformation and attributes it to news outlets. Kohia argued that the lawsuit does not include a single instance in which a consumer confused a legitimate news article with fake AI-generated output.
Mr Justice McMahon said news publishers were under no obligation to do so at this stage and “properly argued there was a risk of consumer confusion”.
Major news organizations including the Globe and Mail and CBC are suing OpenAI in Ontario for copyright infringement. ChatGPT’s creators had argued that Ontario courts lacked jurisdiction and the case should be dismissed. The court rejected that argument last week.
