Skip to main content

Ontario Appeal Court Lets Publishers' Copyright Suit Against OpenAI Proceed

On 13 November 2026 the Ontario Court of Appeal dismissed OpenAI's appeal of a ruling that Ontario courts have jurisdiction over a copyright lawsuit from Canadian news publishers. The Canadian Press, Torstar, The Globe and Mail, Postmedia, and CBC/Radio-Canada say OpenAI trained ChatGPT on their journalism without permission or payment. The case now proceeds in Ontario. The copyright question itself is still undecided.

On this page

The appeal court kept the case in Ontario

The Ontario Court of Appeal on Friday 13 November dismissed OpenAI's attempt to move a copyright lawsuit out of Canada. The company had argued that the case belonged in a United States court. The Ontario Superior Court of Justice had already ruled that it has jurisdiction, and the appeal court's decision leaves that ruling standing, so the case goes ahead in Ontario.

The plaintiffs are a coalition of the country's largest news organizations: The Canadian Press, Torstar, The Globe and Mail, Postmedia, and CBC/Radio-Canada. They filed in late November 2024. They say OpenAI uses their news content to train ChatGPT, and that this breaches copyright and lets the company profit from their work without permission or compensation.

The Canadian Press describes it as the first Canadian case to address the use of copyrighted content to train generative AI systems. Friday's decision is about where the fight happens. It does not decide whether training on news articles is infringement.

What the lower court found about OpenAI's Canadian footprint

The ruling OpenAI appealed is Toronto Star Newspapers Limited v. OpenAI Inc., 2025 ONSC 6217, decided by Justice Kimmel of the Superior Court. The publishers pleaded copyright infringement, unjust enrichment, and breach of contract based on their websites' terms of use. A law firm summary of the motion also lists a claim that OpenAI circumvented technological protection measures.

Justice Kimmel found three connections between the dispute and Ontario that OpenAI did not rebut:

  • Business in Ontario. Some OpenAI entities carry on business in the province through online activity, including crawling and scraping websites, offering services, advertising, and entering contracts.
  • Alleged infringement in Ontario. There was a good arguable case that content was crawled and copied from Ontario and Canadian sites, then used to train and run GPT models offered in Canada.
  • Contracts made in Ontario. There was a good arguable case that the publishers' website terms were binding on OpenAI.

The court kept six OpenAI defendants in the case and dismissed four others. It also held that Ontario was not an inconvenient forum compared with the United States.

Lawyer Barry Sookman, who analysed the decision, noted that it resolved no substantive issue and that OpenAI can still argue at trial that Canada's Copyright Act does not reach conduct that happened outside the country. The published reports of Friday's appeal ruling do not give the appeal court's reasons.

Why a jurisdiction ruling matters beyond Canada

Where an AI copyright case is heard shapes how it ends. United States courts weigh fair use, a flexible defence that AI companies lean on heavily. Canadian law uses fair dealing, which is tied to a list of permitted purposes, and no Canadian court has yet ruled on how it applies to model training. A trial in Ontario means a Canadian judge answers that question under Canadian law.

The reasoning on jurisdiction also travels. The lower court treated crawling a country's websites and selling a service to its residents as enough of a presence to be sued there. If that view holds, an AI developer with no local office can still face each country's copyright rules wherever it scrapes and sells. That is a larger exposure than one lawsuit.

OpenAI could still ask the Supreme Court of Canada for leave to appeal. The reports so far do not say whether it will.

What it means for people who choose and run models

Nothing changes today for anyone using ChatGPT, and no model is being withdrawn. The slower effect is on how training data gets sourced and documented.

Every suit that survives pushes AI developers toward licensed data and clearer records of what went into a model. That applies to open weights too. A model you download and run on your own hardware keeps your prompts private, but it does not tell you whether its training data was licensed. If you build a product on any model, hosted or local, the provenance of its training data is worth checking in the model card and license before you depend on it.

The next milestone is the merits stage in Ontario, where the court will have to look at what OpenAI actually copied and whether any exception covers it.

CuriousLM runs supported AI models locally on your device. Try CuriousLM.