Canadian news publishers accuse Anthropic of bypassing paywalls to feed its Claude AI

The complaint was filed on October 6 in the Ontario Superior Court of Justice, but it only came to light later, through the specialist site CARTT.ca, and was then picked up by The Canadian Press on October 19. The plaintiffs represent most of Canada's print press: Postmedia, Torstar (Toronto Star, Metroland), The Globe and Mail, PNI Maritimes and The Canadian Press. On the other side are Anthropic, the maker of the Claude AI, and its Canadian subsidiary.

The main allegation: Anthropic is said to have copied millions of articles, without a license or payment, to train and run its models. None of these allegations has been examined by a judge. Anthropic did not respond to journalists' requests for comment and has not yet filed a defense.

Bots that allegedly got past paywalls

The complaint goes beyond copyright. According to the summary published by Broadcast Dialogue, the publishers accuse Anthropic's bots, including "ClaudeBot," of bypassing their paywalls and technical protection measures, and of ignoring the terms of use that limit reading to personal use. They add, according to The Canadian Press, that the copying is ongoing, including when Claude fetches articles in real time to answer a user.

The publishers cite very large volumes: about 4.2 million works for The Canadian Press since 2021, 3.5 million for Postmedia and PNI Maritimes, the same for Torstar, and 2.3 million for The Globe and Mail. They also point out that Anthropic has already settled a pirated-books case in the United States for $1.5 billion. In their view, unlike OpenAI, Anthropic has not signed any public licensing agreement with a major news publisher.

What the newspapers want

The plaintiffs are seeking damages for copyright infringement, breach of contract and unjust enrichment, and the disgorgement of profits derived from their content or, failing that, statutory damages of up to $20,000 per work. They also want punitive damages and a permanent injunction barring the use of their articles without written consent.

There is a known precedent: the same publishers, together with CBC/Radio-Canada, have been suing OpenAI since 2024, and that case is still unresolved. The new complaint comes as Anthropic prepares its IPO, while the press attributes spectacular revenue figures to it.

And in Europe?

The question of using content to train AI extends well beyond Canada, as Meta's change to its advertiser terms showed. For website publishers, the case is a practical reminder: the robots.txt file lets you refuse declared AI bots, but it blocks nothing technically. The Canadian complaint, in fact, accuses Anthropic of getting around stronger protections: paywalls and authentication.

Under Ontario rules, a defendant has 20 days to respond if served in Ontario, 40 days elsewhere in Canada or in the United States, and 60 days beyond. Its defense will show whether the company disputes the facts themselves, or only their legal characterization.