Why are Canadian news publishers suing OpenAI?

Large language models learn from enormous amounts of text, and news is some of the best-written, most up-to-date text on the web. That is why publishers on both sides of the border have gone to court. In Canada, a group of news companies started an action in 2024 before the Ontario Superior Court of Justice, on its Commercial List, which handles complex business cases. The plaintiffs are Toronto Star Newspapers, Metroland Media, Postmedia (with PNI Maritimes), The Globe and Mail, The Canadian Press and CBC/Radio-Canada. The defendants are a family of OpenAI companies registered in Delaware and based in California.

Their case is not only about copyright. It rests on three main claims, with a fourth allegation on top, and each one could matter to anyone who publishes online.

What the publishers claim

According to the amended statement of claim, as quoted by the court:

The claim covers content online from 2015 onward. It also says the works may have reached OpenAI through third-party collections such as Common Crawl, a free archive of web pages, and not only through the publishers' own sites.

Which Canadian rules are at stake

Two Canadian legal ideas sit behind these claims. The Copyright Act lets people use works without permission through fair dealing, but only for listed purposes: research, private study, education, parody or satire, plus a few others with conditions. Training a commercial AI model is not on that list as such, so the argument will turn on whether it fits one of them. The Act also forbids circumventing technological protection measures, the digital locks that guard content, subject to narrow exceptions.

What can a publisher check on its own site?

If you publish online yourself, the case is a reminder of the tools the publishers relied on. A worked check you can do this week:

1. Terms of use. Does your site have them, and do they say whether automated copying or commercial reuse is allowed?
2. Crawler rules. Does your robots.txt file address AI training crawlers by name?
3. Access controls. Is paid or members-only content behind a real login, not only hidden from menus?
4. Records. Do you keep dated copies of your terms and robots.txt, so you can show what applied when?

What did the Ontario court decide, and what comes next?

OpenAI's first move was to argue that the case did not belong in Ontario at all. Its companies are American, it said, and the training of its models happened outside Canada, so a Canadian court had no jurisdiction and the United States would be the better place to sue. On November 7, 2025, Justice Kimmel largely rejected that motion in a decision cited as 2025 ONSC 6217.

The decision has three parts:

1. Four OpenAI entities are out. The court stayed the case against four companies, OpenAI's general partner and three startup-fund entities, because they were not shown to be involved in the business at issue.
2. Six stay in. OpenAI, Inc., OpenAI OpCo, OpenAI, LLC and three for-profit parent companies remain defendants. The court found connections to Ontario strong enough to hear the case: these companies carry on business in the province, wrongful acts are alleged to have happened there, and the publishers' Ontario-based terms of use create a contractual link.
3. Ontario is the right forum. OpenAI had not shown that a US court was clearly better placed. The judge added that US courts possibly reaching different results on fair use was no reason to block Canadian claims.

The court was careful to say this was a preliminary ruling and not a decision on the merits. Nobody has yet ruled that OpenAI infringed anyone's copyright in Canada, or that it did not.

What comes next is the slower part. The case now moves toward the merits, where the hard questions will be argued with evidence. One is territorial: Canadian copyright is infringed by acts in Canada, and the judge acknowledged that OpenAI's argument about where the copying took place carries some weight on that point. Others are whether training counts as fair dealing, whether scraping got around protection measures, and whether a crawler can be bound by a website's terms of use. Cases of this kind can take years, and some end in settlement or licensing deals before trial.

Parliament could also change the ground rules. Ottawa held a public consultation on copyright and generative AI that closed in January 2024 and published a "What We Heard" report. Its stated next step is that the government continues to consider how to address these concerns in the copyright framework or by other means. As of October 2026, no amendment covering AI training has been adopted.

For anyone who simply uses ChatGPT, nothing changes today; the service keeps running while the case proceeds. If you run a publication and are weighing your own claim or a licensing offer, that is a conversation for a copyright lawyer, since the answers depend on what you own and where the copying happened.