Executive Summary: Key Legal and Evidentiary Issues

• Plaintiffs allege OpenAI infringed their copyright, breached terms of use, and bypassed digital protections to train AI models.

• OpenAI brought a jurisdiction motion, asserting Ontario courts lack authority over its entities and operations.

• Defendants sought a partial sealing order to prevent disclosure of confidential commercial and technical information.

• The court applied the Sierra Club and Sherman Estate test, focusing on necessity, proportionality, and public interest.

• Evidence showed a real and substantial risk of competitive harm if sensitive AI infrastructure details were disclosed.

• The court emphasized that procedural fairness in highly competitive sectors includes protecting trade secrets from public exposure.

Factual background and procedural history

This case arises from a civil claim filed by a coalition of major Canadian media organizations, including Toronto Star Newspapers Limited, the CBC, and The Globe and Mail. The plaintiffs allege that OpenAI unlawfully used their copyrighted content to train AI systems such as ChatGPT, in violation of Canadian copyright law, their websites’ terms of use, and technological protection measures meant to prevent automated content scraping.

The defendants are a group of U.S.-based companies affiliated with OpenAI. Without conceding to the court’s authority, they brought a jurisdiction motion arguing that Ontario courts lack jurisdiction over them or the conduct alleged. Before that motion could be heard, the defendants moved for a partial sealing order to protect certain documents and testimony containing confidential business information that they planned to use in the jurisdiction motion.

The motion for a partial sealing order, which is the subject of this decision, was heard by Justice Kimmel on July 30, 2025. The ruling focuses on whether the court should allow portions of the defendants’ evidence to be shielded from public access to avoid harming their competitive position in the rapidly evolving AI industry.

Nature of the confidential information and parties’ positions

The information the defendants sought to seal included:

• Internal corporate structure, including detailed organizational charts

• Technical specifics of web crawling, fetching, and data infrastructure

• AI training and inference systems and their geographic and operational footprints

• Cost structures and proprietary methods related to data access and model training

• Licensing arrangements and intellectual property holding patterns across OpenAI entities

OpenAI filed affidavits, particularly from James Sherwood (Legal Staff, OpenAI OpCo), asserting the commercial sensitivity of the information and detailing how disclosure could cause harm by allowing competitors—such as Google, Amazon, or AI startups—to reverse-engineer their practices or capitalize on strategic business information.

The plaintiffs did not oppose the sealing order outright but urged the court to ensure that OpenAI met the legal burden of proof. Meanwhile, non-party journalists from the Toronto Star and CBC formally opposed the order, arguing that transparency and the open court principle were paramount, especially in a case involving public interest and media rights.

Application of the legal test: necessity and proportionality

Justice Kimmel reviewed the motion under the well-established test for confidentiality orders articulated in Sierra Club of Canada v. Canada (Minister of Finance) and later clarified in Sherman Estate v. Donovan. The test requires three elements:

1. A serious risk to an important public interest that would result from openness;

2. Necessity of the order to prevent that risk because no reasonable alternatives exist;

3. Proportionality, meaning the benefits of the order outweigh its harmful effects on public court access and expression.

On the first branch, the court accepted OpenAI’s evidence that disclosure posed a real and substantial risk to the important public interest of fair competition. Justice Kimmel emphasized that this interest extended beyond OpenAI’s private interests—it implicated the public’s stake in a marketplace that rewards innovation rather than exposing confidential strategies through litigation. The court acknowledged the heightened sensitivity of commercial information in the AI space, where rapid advancement and secrecy are often crucial to competitive survival.

Under the second branch, the court found the proposed order to be narrowly tailored. OpenAI had only designated specific excerpts and one document for sealing. These redactions were necessary to enable OpenAI to defend itself without compromising its competitive position. Alternatives, such as summaries or additional redactions, were deemed insufficient.

On proportionality, the court concluded that the public interest in disclosure was minimal in comparison to the risk of competitive harm. The sealed information did not pertain to matters of political participation, public accountability, or personal rights. Further, the court noted that the public would still have access to the bulk of the record, and that designated representatives of the plaintiffs could review the sealed content under protective conditions. This limited the impact on the open court principle.

Doctrine of the case: balancing commercial confidentiality and open justice

This ruling reinforces a critical doctrine in Canadian jurisprudence: the qualified nature of the open court principle, particularly in commercial and technological disputes. Courts have long recognized that while judicial proceedings must generally be transparent, exceptions are warranted where disclosure would undermine other important public interests—such as encouraging fair economic competition or protecting proprietary innovation.

The court reiterated the doctrinal principle that “not all commercially sensitive information must be formally copyrighted, patented, or memorialized in technical manuals” to qualify for sealing. Commercial confidentiality is a broader concept, especially in sectors like artificial intelligence, where competitive advantage often hinges on operational models and resource deployment strategies rather than formal intellectual property rights.

Importantly, the court distinguished between the private interest of shielding embarrassing information and the public interest in maintaining a fair and competitive market. In doing so, it emphasized that litigants should not be forced to choose between asserting legal rights and protecting legitimate business secrets.

Outcome and implications

The court granted the defendants’ motion for a partial sealing order. It ruled that the evidence satisfied the high legal threshold under the Sherman Estate test and confirmed that the order was both necessary and proportionate.

While the plaintiffs and journalists raised valid concerns about transparency, the court found that the nature of the information, the narrow scope of the sealing request, and the public interest in preserving competitive integrity justified the order.

Justice Kimmel emphasized that the sealing order applied only to the jurisdiction motion and would not necessarily extend to later phases of the litigation. Future requests for confidentiality would need to be justified independently.

This decision reinforces the growing recognition that in cases involving advanced technologies and AI, courts must carefully balance transparency with the need to protect innovation and preserve fair competition, especially when the litigating party did not initiate the lawsuit and is seeking only to defend its jurisdictional position.