A bill for 13.5 million articles
On October 6, 2026, Canada’s largest news publishers walked into the Ontario Superior Court of Justice and handed Anthropic the bill for Claude’s education. The coalition — Torstar, The Globe and Mail, Postmedia Network and The Canadian Press — alleges the AI company systematically harvested their journalism to train and operate Claude, and it has named a price: the statutory maximum of C$20,000 per infringed work.
The publishers’ own count puts the haul at roughly 13.5 million works taken since 2021: about 4.2 million from The Canadian Press, 3.5 million each from Postmedia and Torstar, and 2.3 million from The Globe and Mail. Do the arithmetic and you get C$270 billion. That figure is the arithmetic result of multiplying the publishers’ approximate works count by the requested statutory maximum — a ceiling the publishers are asking the court to recognize, not a prediction of what a court would eventually award. A demand that reads, plausibly but not provably, like an opening bid: set the ceiling high enough and whatever Anthropic eventually offers looks like a bargain.
That last sentence is an inference, not a finding. But the evidence for it is hiding in plain sight, in the publishers’ own statement of claim.
The hint buried in the complaint
The claim accuses Anthropic of deliberately scraping the publishers’ journalism — including via its ClaudeBot crawler — in breach of website terms of use and past paywalls and login screens, without licence or compensation. On top of the statutory damages it seeks punitive damages for wilful conduct and a permanent injunction stopping Anthropic from using the publishers’ current and future works without written consent.
Then comes the interesting part. The publishers go out of their way to contrast Anthropic with “other AI companies” that “have engaged meaningfully with publishers in other jurisdictions, resulting in legitimate market-based licensing solutions that fairly compensate content creators” — accusing Anthropic, in the claim’s telling, of choosing to “brazenly misappropriate” instead.
Read that again. A complaint whose headline relief runs to hundreds of billions of dollars takes a full paragraph to say, in the publishers’ own telling: other AI companies cut licensing deals in other jurisdictions; you didn’t. That is suggestive, not proof, of a plaintiff with an interest in a deal — of a plaintiff pricing one. The filing never says the publishers would settle, and whether they would is still an open question.
Anthropic has paid before
In 2025, Anthropic agreed to pay US$1.5 billion to settle a US author class action over books downloaded from pirate sites, after a judge ruled that training on legally bought books was fair use but keeping pirated copies was not. That is the factual record: one very large settlement, agreed after a mixed court ruling — not a demonstrated habit. Whether exposure at that scale is what moved Anthropic then, or would move it in Ontario, is inference, not evidence.
One plausible reading of the Ontario number is that it was set with that history in mind: put the ceiling at C$270 billion and any licensing figure Anthropic brings to the table — even a very large one — arrives looking like mercy. Litigation as leverage, conducted in the language of damages — if the publishers are open to a deal at all, which nothing in the filing confirms.
Unless they mean it
Fairness requires the steelman, and it is a strong one. Canada is not the United States: there is no broad fair-use defence here, only the narrower fair-dealing doctrine — a thinner shield for AI training than Anthropic enjoyed south of the border. The permanent injunction the publishers seek, backed by the paywall-and-login circumvention claims, is not decorative. It has real legal teeth.
And these publishers are not first-time bluffers. The same coalition, joined by CBC/Radio-Canada, sued OpenAI in the same court in November 2024 — and that case still hasn’t settled, nearly two years on. A group still litigating one frontier AI company nearly two years on may be perfectly prepared to do it twice. Reading the Anthropic suit purely as leverage could underestimate a coalition whose OpenAI case is still alive — suggestive of resolve, not yet proof of a willingness to pay the price of a ruling.
Then there is Anthropic’s side of the ledger. Broadcast Dialogue reports the company had raised roughly US$125 billion and was valued at more than US$965 billion, adding over a million new Claude users a day as of May. A company that rich might prefer a precedent-setting legal win to paying Canadian newsrooms a toll — and a loss in Ontario would be far worse for the publishers than a bad deal. How long each side can afford to find out is a question the public record doesn’t answer.
What happens next
For now, mostly waiting. Anthropic had not responded to requests for comment from Cartt.ca or The Canadian Press, and whether it has filed a statement of defence remains unresolved. None of the allegations have been proven in court.
The questions that matter are all still open. Will Anthropic argue fair dealing, and how will an Ontario court treat AI training under Canadian copyright law? Are the publishers open to a licensing deal at a concrete price, or do they genuinely want the ruling and the injunction? Did ClaudeBot actually defeat paywalls and logins, or is the circumvention claim a stretch of a terms-of-service argument? How would a permanent injunction over “future works” even be enforced — a Canadian-journalism filter bolted onto Claude? And does Anthropic’s new Canadian data centre in Nova Scotia factor into the leverage at all?
The publishers have named their price. The number that matters is the one nobody has named yet: the price at which this stops being a lawsuit and starts being a licensing deal. If there is one.
