Publications

By: Douglas Cottier & Heather McCann

In a significant decision at the intersection of artificial intelligence and copyright law, the Ontario Superior Court of Justice (the “Court“) in Toronto Star Newspapers Limited v. OpenAI Inc., 2025 ONSC 6217 (“TSNL“) , confirmed that it may assume jurisdiction over claims alleging that OpenAI Inc. and its related entities (collectively, “OpenAI“) unlawfully scraped copyrighted materials from Canadian news organizations to train its large language models (“LLMs“). While the Court did not address the merits of the claim, the decision provides guidance on how jurisdictional principles may apply to cross-boarder AI-related disputes.

What are large language models and how do they leverage data scraping?

Before turning to the Court’s analysis, it is helpful to understand concepts such as LLMs and “data scraping.”

LLMs are AI systems trained and developed through large volumes of text-based data, commonly referred to as “datasets”. Datasets enable LLMs to identify patterns in language, allowing them to generate their “human-like” responses. As a result, LLMs require extensive datasets to function.

In some cases, these datasets are created through “scraping.” Scraping is an automated process which involves the use of web crawlers or bots to extract information from various online sources. The extracted information is then incorporated into datasets used to train LLMs. Once fully developed, LLMs go on to function as the underlying engines powering AI platforms such as ChatGPT. ¹

What facts and issues were before the Court in TSNL?

The plaintiffs consisted of several Canadian news organizations, including the Toronto Star, the Globe and Mail, and the Canadian Broadcasting Corporation. They allege that OpenAI copied their online content without authorization and subsequently incorporated it into datasets used to train various LLMs. The plaintiffs further alleged that once trained, these LLMs were built into AI platforms such as ChatGPT, which were made available to users in Canada.

OpenAI moved to stay or dismiss the proceeding, arguing that the alleged conduct occurred outside Canada and as such an Ontario court lacked jurisdiction.

The Court went on to address three central issues:
(a) whether it had subject matter jurisdiction over OpenAI;
(b) whether it had in personam jurisdiction over OpenAI; and
(c) whether it should decline jurisdiction on the basis of forum non conveniens.

What is subject matter jurisdiction and in personam jurisdiction?

Subject matter jurisdiction is a court’s authority to hear and decide a specific type of case whereas in personam jurisdiction refers to the court’s authority to make a decision that binds a person or entity.

The Court held that it had subject matter jurisdiction over all asserted claims. As a court of general jurisdiction, it may hear any claim unless that authority is clearly removed by statute or agreement, neither of which was present in this case.

OpenAI raised arguments regarding the territorial limits of copyright and the scope of the Copyright Act which the Court noted do not bear on the questions of subject matter jurisdiction. The Court likewise rejected OpenAI’s position that the Copyright Act operates as a complete code that precludes parallel claims in contract and unjust enrichment, holding that these are merits-based arguments rather than jurisdictional ones.

Turning to in personam jurisdiction, a court may assume jurisdiction over an out-of-province defendant where there is a “real and substantial connection” between the subject matter of the dispute, the parties, and the location that the dispute will be heard.

Following the two-step test developed in Club Resorts Ltd. v. Van Breda, 2012 SCC 17, a plaintiff will establish a connection to a province by identifying one or more “presumptive factors.” ² Such a factor may arise, for example, where there is a contract made in Ontario that is connected to the dispute in question. The burden then shifts to the defendant to rebut the presumption by demonstrating that the connection is insufficient to establish a real relationship between the dispute and the forum.

Ultimately, the Court in TSNL held that it had in personam jurisdiction over six of the ten defendants. In making this finding, the Court accepted several presumptive connecting factors, including that the plaintiffs’ Terms of Use constituted contracts connected to Ontario and that such would be breached through OpenAI’s alleged scraping activities.

The defendants were unable to rebut the strength of those connections.

What is forum non conveniens and how was it decided in TSNL?

Forum non conveniens permits a court with jurisdiction to decline to exercise it where another forum is clearly more appropriate to hear the dispute.

The defendants in TSNL, having raised forum non conveniens, bore the burden of establishing that their proposed forum (the United States) was more appropriate to adjudicate the dispute.

In making an assessment under forum non conveniens, a court will consider a non-exhaustive list of factors which include but are not limited to the sophistication of the parties, the relevant law to apply to the dispute, and the enforcement of an eventual judgment.

While several non‑exhaustive factors were discussed in TSNL, the Court’s observation that, in the interest of fair and efficient administration of justice, Canadian authors should be able to pursue claims in Canada against foreign defendants when those defendants are properly subject to a particular Canadian court’s jurisdiction. Ultimately, the Court rejected the defendants’ position that the United States was a clearly more appropriate forum.

What was the outcome and what comes next?

In the result, the action was stayed as against four defendants ³ and will proceed against the remaining six, namely, OpenAI Inc., OpenAI OpCo LLC, OpenAI LLC, OpenAI Holdings LLC, OAI Corporation, and OpenAI Global LLC.

Although the unsuccessful defendants have appealed the Court’s decision, should the decision withstand an appeal, it will represent an interesting development in Canadian jurisprudence as courts rely on this decision to rule on the substantive question of whether the use of copyrighted materials in training AI systems gives rise to liability under Canadian copyright law.

If you have any questions regarding this article, please contact the authors, Douglas Cottier or Heather McCann or any member of our Business Law group.

[1] For more information on LLMs and scraping see the following sources: Humza Naveed et al, “A Comprehensive Overview of Large Language Models” (2023); Enrique Ayuso et al, “From manual to machine: How AI is redefining web scraping for superior efficiency: A literature review.” (2024).

[2] Presumptive factors may include (a) the defendant is domiciled or resident in Ontario; (b) the defendant carries on business in Ontario; (c) the tort was committed in Ontario; (d) a contract connected with the dispute was made in Ontario; and (e) property relating to the asserted claims is located in Ontario.

[3] OpenAI GP LLC, OpenAI Startup Fund Management LLC, OpenAI Startup Fund 1 LP, and OpenAI Startup Fund GP 1 LLC.