A filing from 6 October, a story from 19 October
Start with the dates, because coverage this week has blurred them. The statement of claim was filed on 6 October. Cartt.ca reported it first, and trade outlet Broadcast Dialogue published the detailed breakdown on 14 October. The story reached a wider audience on 19 October, when The Canadian Press confirmed the case and Canadian outlets picked it up. Nothing about the lawsuit itself is new this morning. What changed is that Canada's national wire service put its name to it. One disclosure belongs up front: The Canadian Press is also one of the plaintiffs, so its role as both a party to the case and the confirming wire service is relevant context for readers weighing the coverage that followed.
The plaintiffs are the core of English-language Canadian news publishing: Torstar, owner of the Toronto Star; The Globe and Mail; Postmedia, the country's largest newspaper chain; and The Canadian Press, the wire whose copy appears in outlets nationwide. Together, by their own count in the claim, they hold or license about 13.5 million works published since 2021. That total mixes two kinds of rights: it includes works the plaintiffs say they own and works they license from others, in proportions that differ by publisher. The claim therefore cannot be reduced to a simple count of publisher-owned articles.
Training is only half the claim
The familiar part is the training allegation: that Anthropic "deliberately 'scrapes'" enormous amounts of text from the publishers' websites and uses it, without authorisation, to develop its models. The claim adds three charges that go further. First, breach of contract, on the basis that crawlers such as ClaudeBot ignored site terms limiting content to personal use. Second, circumvention, the allegation that Anthropic bypassed paywalls, login requirements and network protections built to stop copying. Third, and most consequential for how assistants operate now, the claim says the scraping is ongoing, driven by user prompts. That allegation extends the claim beyond past model training to alleged ongoing retrieval and summarisation. Whether Anthropic engaged in the conduct described, and whether it infringed the plaintiffs' rights, remains to be determined.
The publishers also make a market argument. Claude, they say, returns their work verbatim, paraphrased or summarised, which gives readers a substitute for visiting the source, and that substitution costs subscriptions, advertising and licensing revenue. Their sharpest line is comparative, and it is the plaintiffs' argument, not a court's finding: rival AI companies have signed public licensing deals with publishers, while Anthropic, in the claim's words, "has yet to execute any form of public content licensing agreement with major publishers" and has "required litigation and Court intervention as a precondition to paying content creators anything." Those quoted lines are the claim's characterisations of the licensing market and of Anthropic's conduct. The claim invokes Anthropic's US books litigation in support, describing it as involving more than seven million pirated books and pointing to a US$1.5 billion settlement. That is the plaintiffs' framing of the US case, not a finding that all AI training on copyrighted books is unlawful: the US litigation addressed distinct issues, including the fair use of lawfully acquired books and, separately, the acquisition of pirated copies.
Do the arithmetic, then discount it
The relief section is where the claim stops sounding like a complaint and starts sounding like a price list. The publishers ask for general damages and an accounting of profits, or, as an alternative to proving actual losses, statutory damages of up to C$20,000 per infringed work, plus punitive and exemplary damages and a permanent injunction covering current and future works.
Multiplying the claim's two headline numbers — about 13.5 million works and a statutory maximum of C$20,000 per work — produces C$270 billion. That is our arithmetic, not a damages estimate supplied by the plaintiffs, and it should not be read as the amount a court would award. Statutory damages are a legal remedy, not an automatic tariff: a court would have to determine the applicable basis and amount, and the claim itself pleads alternative forms of relief. Broadcast Dialogue's description of the potential exposure as "potentially billions" is the more cautious indication of scale, and even that rests on allegations Anthropic has not yet answered. The same caution applies to the claim's portrait of Anthropic's finances: roughly US$125 billion raised and a valuation above US$965 billion as of May, with over a million people signing up for Claude per day. Those figures appear in the pleading to support an unjust-enrichment argument. They are the plaintiffs' assertions, not audited numbers.
The injunction request may prove as consequential as the damages claim, but it is still a request, not an order. The publishers seek a permanent injunction barring Anthropic from accessing or using their current and future works without written consent. If granted in that form, it could materially change the licensing position between the parties, forcing the conversation the publishers say they have been denied, on their timetable rather than Anthropic's. Whether the court grants such relief, and on what terms, remains unresolved.
What happens next
Anthropic has up to 60 days from service outside Canada to file a defence, and it offered no comment to Cartt or The Canadian Press. The publishers' parallel case against OpenAI, filed with CBC/Radio-Canada in November 2024, is a separate proceeding; it remains unsettled after nearly two years, which sets expectations for pace here without deciding anything in this one. The publishers have also spent that period asking Ottawa to address AI training in federal policy, after the national AI strategy left the copyright question largely untouched. This filing reads as the courtroom version of that lobbying: if government will not set a price for training data, a judge will be asked to.
Sources
- [1] Broadcast Dialogue, "Canadian news publishers sue Anthropic over AI copyright infringement", published 14 October 2026, updated 15 October 2026 — plaintiffs and works table, four heads of claim, relief sought including C$20,000 per work and injunctions, the claim's Anthropic fundraising and valuation figures, the US books-case citation, and the 60-day defence windowRead source
- [2] MobileSyrup, "Canadian publishers launch copyright suit against Anthropic", published 19 October 2026 (sourcing Global News / The Canadian Press) — filing date of 6 October, Cartt as first reporter, the quoted unjust-enrichment passage, the ongoing-scraping allegation, and the still-unsettled 2024 OpenAI caseRead source
- [3] Mad Robot (summarising Cartt.ca and The Canadian Press reporting), "Globe, Star, Postmedia and CP sue Anthropic over Claude training", 19 October 2026 — the four claim heads in plain form and confirmation that Anthropic did not respond to Cartt or CP and has filed no defenceRead source
- [4] The Neuron, "Everything That Happened in AI Today — Tuesday, October 20, 2026" — the suit listed among the day's top five developments, with the plaintiff coalition describedRead source