Who owns AI-generated content in Canada

In 2026, AI continues to test the limits of Canadian intellectual property law. Who owns the outputs of AI tools, and may copyright-protected works be used as inputs to train them? Canadian law offers no clear answer. With Canadian case law and the statutory framework still underdeveloped, businesses continue to look to foreign courts and regulators for guidance.

That guidance is far from settled. In 2025, a regional German court held that OpenAI infringed copyright by training its model on song lyrics: the model memorized the works and produced near-identical outputs. By contrast, the U.K. High Court found that Stability AI’s model did not infringe Getty Images’ copyright because it neither stored nor reproduced the protected works.

Closer to home, AI is also blurring traditional notions of human inventorship. Here, at least, Canada has drawn a line. In Thaler, Stephen L. (Re), 2025 CACP 8, the Patent Appeal Board confirmed that only natural persons can be inventors under Canadian law and rejected Thaler’s patent applications for inventions created by his AI system, DABUS.

The question of copyright ownership is murkier. Canadian copyright law still requires a human author, but measuring human contribution to AI‑assisted work is increasingly difficult. In 2024, the Canadian Intellectual Property Office (CIPO) issued a copyright registration for Suryast, an artwork credited to an individual and the AI software RAGHAV. That same year, petitioners challenged the registration in the Federal Court, arguing that AI authorship lacks both originality and a human author.

The United States has taken a firmer stance. The U.S. Copyright Office has refused to register works generated solely by AI, a position supported by the courts: in Thaler v. Perlmutter, the D.C. Circuit held that copyright requires a human author, and the Supreme Court denied certiorari on March 2, 2026.

Canadian copyright law has yet to provide the same clarity. Until the courts and Parliament provide further guidance, businesses must manage the risks through contracts and internal policies.

Risks for businesses: operating in a legal grey zone

This absence of clarity creates several practical risks for Canadian businesses using generative AI commercially.

1. Copyrightability of AI-generated content

AI-generated content may not attract copyright protection at all. If a Canadian court follows the Supreme Court of Canada’s decision in CCH Canadian Ltd. v. Law Society of Upper Canada, 2004 SCC 13 (CCH) and the emerging international consensus that authorship requires human creativity, purely AI-generated outputs, including marketing copy, design concepts, code, reports, may not qualify for copyright protection, leaving businesses with no legal recourse under copyright law if competitors copy them. Even AI-assisted works, in which a human provides creative direction but AI generates the final expression, occupy uncertain territory: the degree of human involvement required to establish copyright protection has not been tested in Canadian courts.

2. Training data and third-party infringement exposure

Generative AI models are trained on vast datasets that may include copyrighted material, and their outputs may reproduce substantial portions of those works, exposing users to infringement claims. A coalition of Canadian news media organizations, including the Toronto Star, The Globe and Mail and CBC/Radio-Canada, has sued OpenAI over the use of its content to train ChatGPT. In November 2025, the Ontario Superior Court affirmed its jurisdiction, finding a “good arguable case” that copyright infringements had occurred in the province. That decision is now under appeal.^[1] Six additional proposed class actions have been commenced in the Federal Court and in Québec and British Columbia. Commercial use of AI-generated content may therefore expose businesses to downstream infringement risks arising from a vendor’s training practices, even when they have no knowledge of or involvement in those practices.

3. Ownership disputes: employer, employee and vendor

Under section 13(3) of the Copyright Act, employers are generally the first owners of copyright in the works created by employees in the course of their employment.^[2] However, if an employee’s AI-generated output lacks human authorship, there may be no copyright for the employer to own. If the output qualifies for copyright protection, perhaps because the employee exercised sufficient creative control, questions of authorship and ownership may arise among the employee who wrote the prompt, the employer who directed the project and the AI vendor who developed the model. Similarly, a vendor’s licence to use AI-generated outputs is only as valuable as the underlying rights it conveys and may offer limited protection if no human author exists.

Contractual and policy solutions: what businesses can do now

In the absence of legislative reform, Canadian businesses must manage these risks through contractual protections and internal policies. While the following measures will not eliminate the underlying uncertainty, they can significantly reduce potential exposure.

1. IP assignment and ownership clauses

Agreements with AI vendors, employees and contractors should include clear IP assignment language. Vendor agreements should specify who owns platform outputs and whether the vendor retains any rights, including the right to use those outputs for model training. Employment agreements should assign to the employer all work product created using AI tools in the course of employment, regardless of the degree of AI involvement. These provisions cannot create copyright where none exists, but they allocate contractual rights and reduce the risk of ownership disputes.

2. Representations, warranties and indemnification

Businesses procuring AI services should require vendors to represent and warrant that their models do not infringe third-party intellectual property rights and that their outputs will not expose users to infringement liability. These protections should be supported by an indemnity requiring vendors to defend and hold users harmless against third-party infringement claims arising from the platform or its content. Vendors should also disclose the nature and provenance of their training data to the extent commercially practicable. The Government of Canada’s consultation report noted “significant interest in developing transparency requirements surrounding the use of copyright-protected works in the training of AI.”^[3]

3. Internal AI use policies

The most immediately actionable step is to establish an internal AI use policy that, at a minimum, sets guidelines for when and how employees may use generative AI tools. The policy should require employees to document their creative contributions to AI-assisted output, so the organization can demonstrate the “skill and judgment” needed to support a copyright claim under CCH. It should also prohibit employees from inputting confidential, proprietary or third-party copyrighted material into external AI platforms without authorization and designate responsibility for reviewing outputs before commercial use. Documentation is particularly important. In January 2025, the U.S. Copyright Office registered A Single Piece of American Cheese, an image created through 35 rounds of modification using AI inpainting tools, based on demonstrated iterative human creative control.^[4] Canada has no equivalent guidance, but the stronger the record of human creative involvement, the stronger the case for copyright protection.

The case for acting before the law catches up

Canada stands at a crossroads. The foundational concepts of the Copyright Act (authorship, originality, skill and judgment) do not map neatly onto the outputs of generative AI. CIPPIC v. Sahni may provide the first judicial guidance, but a single Federal Court decision cannot substitute for the comprehensive legislative reform this moment demands.

Other jurisdictions are moving forward. Through Thaler v. Perlmutter and the Copyright Office’s multi-part report^[5], the United States has reinforced the requirement for human authorship while leaving open the possibility of copyright protection for AI-assisted works involving sufficient human creative control. The United Kingdom is preparing to repeal its computer-generated works provision, and the European Union has built transparency and opt-out mechanisms into its regulatory architecture.

Canada, by contrast, has yet to establish a comprehensive legislative or regulatory framework addressing AI-generated content, and judicial guidance remains limited. Meanwhile, businesses are deploying generative AI at scale, creating work product with uncertain legal status and exposing themselves to risks they may not fully appreciate.

For counsel advising businesses on corporate, technology and data privacy matters, the prudent course is to act as though that future has already arrived. This means drafting AI vendor agreements that allocate output rights and establish training-data warranties and indemnities with precision; updating employment and contractor agreements to capture AI-assisted work product; and implementing internal governance policies that document human creative contribution as rigorously as organizations document their handling of personal information. The businesses that treat provenance, transparency and human authorship as compliance disciplines today will be best positioned when the courts and Parliament speak. The question of who owns AI-generated content is not merely academic. Its answer will shape creativity, commerce and intellectual property in this country for decades to come.

***

Roland Hung is a partner, Chair of the Privacy, Data & Cybersecurity Group and Co-chair of the AI & Innovation Group. His business law practice focuses on technology, privacy compliance, cybersecurity and data management. Drawing on his experience as former Senior Legal Counsel and Global Privacy Officer at a large multinational, and Chief Legal Officer and Chief Privacy Officer of a technology company, Roland provides practical advice to organizations ranging from financial institutions to not-for-profits, on matters including artificial intelligence and other emerging technologies. He represents clients before privacy regulators, courts and administrative boards. Roland is ranked by Chambers Canada, Chambers Global and The Canadian Legal Lexpert Directory.