Four Canadian publishers sue Anthropic over Claude training

Four Canadian news publishers have sued Anthropic in Ontario, alleging that the company copied their journalism without permission to develop Claude, its large-language AI model. The claim was filed Oct. 6 by Postmedia, Torstar, The Globe and Mail and The Canadian Press. The publishers allege that Anthropic scraped enormous quantities of text from the internet and their sites for its models and services, in breach of website terms of service and technological measures intended to limit copying. They also allege that the company continues to scrape material in response to user prompts. The allegations have not been tested in court, and Anthropic had not responded to requests for comment when the reports were published. The statement of claim says Anthropic built and commercialized an expanding suite of AI services without a valid licence from any of the news companies. The four plaintiffs say they produce a significant share of Canadian journalism and collectively reach millions of readers. The action is the same group of publishers' second major AI copyright case. They and CBC/Radio-Canada sued OpenAI in 2024 in a case that remains unresolved. As covered in July, Anthropic separately reached a $1.5 billion settlement in a U.S. dispute over books used to train Claude. Press Gazette described the Canadian case as the first lawsuit by publishers against Anthropic in its tracker, following Reddit's June 2025 action. Reddit alleged that Anthropic had accessed its site more than 100,000 times in the preceding year despite requests that its bots stop. The Canadian suit arrives as publishers pursue a mixture of licensing arrangements and litigation over AI companies' use of copyrighted material. They are also concerned that readers may get answers from chatbots without clicking through to the original reporting, weakening traffic and advertising income. The U.S. Copyright Office has said generative AI relies on vast data collections that include copyrighted works, and identified data collection, training, retrieval-augmented generation and model outputs as points where copyright issues can arise. Its review noted dozens of pending U.S. cases examining whether AI training qualifies as fair use, while considering potential licensing and opt-out systems. Canadian publishers have also pressed the federal government to clarify how AI systems may use copyrighted material. According to The Globe and Mail, Canada’s national AI strategy had not addressed that question.