Skip to content
Intellectual Property

AI-Generated Works and Copyright Ownership in Canada

AI-generated works are protected by Canadian copyright only to the extent they reflect a human author's skill and judgment, under the originality test in CCH (2004 SCC 13). The Copyright Act has no AI-specific rules, so businesses must secure ownership through contracts and documentation.

Digital artwork created with generative AI, illustrating copyright ownership of AI-generated works in Canada

Key takeaways

  • Originality requires a human exercise of skill and judgment that is more than mechanical.
  • The Copyright Act, current to 19 October 2026, contains no AI-specific provisions.
  • Registrations naming AI as an author do not settle whether copyright exists.
  • Contracts with AI providers, staff and contractors should address ownership of outputs.
  • Deepfakes raise privacy, personality and advertising issues beyond copyright.

AI-generated works and copyright in Canada: the short answer

Whether AI-generated works are protected by copyright in Canada is unsettled, but the existing law points to a clear working rule: copyright protects original expression that results from a human author's exercise of skill and judgment. The Copyright Act protects "original" literary, dramatic, musical and artistic works, and in CCH Canadian Ltd. v. Law Society of Upper Canada, 2004 SCC 13, the Supreme Court of Canada held that originality requires an exercise of skill and judgment that is more than a purely mechanical exercise.

Output produced autonomously by an AI system, with little human creative input, is unlikely to meet that test. Output in which a human makes meaningful creative choices, by selecting, arranging, editing or combining AI-assisted material, is more likely to be protected, at least as to the human contribution.

Digital artwork created with generative AI, illustrating copyright ownership of AI-generated works in Canada
Copyright depends on human skill and judgment, not on the tool used to create a work.

As at the consolidation current to 19 October 2026, the Copyright Act contains no provision dealing specifically with artificial intelligence. The federal government consulted on copyright and generative AI, with submissions closing on 15 January 2024, and published a report on what it heard, but Parliament has not amended the Act. Businesses therefore need to manage ownership of AI-generated works through contracts and documentation. This article explains the originality test, authorship, registrations, contractual ownership, deepfakes and compilations.

Originality and the human skill and judgment test

In CCH (SCC docket 29320), the Supreme Court rejected both a pure "sweat of the brow" test and a requirement of creativity. It held that an original work must be the product of an author's exercise of skill and judgment: skill being the use of knowledge, developed aptitude or practised ability, and judgment being the use of one's capacity for discernment or ability to form an opinion by comparing different possible options. The exercise cannot be so trivial that it could be characterized as purely mechanical.

Applied to generative AI, the question is where the skill and judgment lie. Typing a short prompt and accepting the first output involves little authorial choice in the expression itself. Iterative prompting, detailed instructions about composition, selection among many outputs, and substantial human editing involve more. Courts have not yet ruled on where the line falls, and the answer is likely to depend on the facts of each case and the type of work.

ScenarioLikely position (indicative)Why
Fully autonomous output with a short promptWeak or no copyrightLittle human skill and judgment in the expression
Human selects and arranges many AI outputs into a larger workCopyright in the selection and arrangement may existCompilation protects original selection or arrangement
Human substantially edits or adds to AI outputCopyright in the human contributionsOriginal human expression is identifiable
AI used as a tool within a human-led creative process (e.g., editing software features)Generally protectedThe human remains the author making the creative choices
Code generated by AI and integrated by developersMixed; human-written and human-selected parts protectedDepends on extent of human authorship

Authorship of AI-generated works and the term of protection

The Copyright Act is built around human authors. Copyright subsists under s. 5(1) where, among other conditions, the author was a citizen, subject or ordinary resident of a treaty country when the work was made, and the general term of copyright is the life of the author plus 70 years (s. 6). Those provisions assume a natural person whose nationality, residence and death can be identified. Moral rights under s. 14.1, such as the right to integrity and attribution, also attach to authors. An AI system has no nationality or lifespan and cannot hold rights, which supports the view that it cannot be an author under the Act.

That does not mean the developer of the AI system or its user automatically becomes the author. Authorship follows the person who exercised skill and judgment in the expression. A developer who built the model but had no role in a particular output is unlikely to be its author; a user who directed and shaped the output may be.

Copyright registrations naming AI

Canada's copyright registration system is a registration system, not an examination system: the Copyright Office generally records applications without assessing whether copyright actually subsists. As a result, registrations listing an AI tool as an author or co-author can be recorded. Under s. 53(2), a certificate of registration is evidence that copyright subsists and that the registered person is the owner, but that presumption can be rebutted. A registration naming an AI system therefore does not settle whether the work is protected, and it may invite a challenge.

Practical point

Document human contributions to important AI-assisted works: prompts, drafts, selections, edits and decisions, with dates and names. That record is the best evidence of originality and authorship if ownership is challenged, and it supports registration that names the human authors. Keep a dated record of the analysis, the documents relied on and who approved each decision, so that the reasoning can be explained later if questions arise.

Securing ownership of AI-generated works through contracts

Because the statutory position is uncertain, contracts carry most of the weight. Businesses using generative AI should address four layers:

  1. AI provider terms: check who owns outputs under the tool's terms of service, what licences the provider takes in prompts and outputs, whether the provider offers indemnities for third-party claims, and what confidentiality applies to inputs.
  2. Employees: employment agreements and policies should confirm that AI-assisted work made in the course of employment belongs to the employer, and that employees follow approved tools and disclosure rules.
  3. Contractors and agencies: require disclosure of AI use, assignment of all rights in deliverables, warranties about third-party content, and moral rights waivers; see our insight on IP ownership between employees and contractors.
  4. Customers and licensees: state what rights are granted in deliverables that may include unprotected AI output, and limit warranties of exclusivity where copyright may not subsist.

Contractual allocation works between the parties, but it cannot create copyright where none exists. Where outputs may not be protected, businesses should rely on confidentiality, trade secrets, trademarks and speed to market for commercial protection.

Infringement risk in AI outputs

AI-generated works can also infringe. If an output reproduces a substantial part of a protected work, the person who reproduces or distributes it may be liable, whether or not the reproduction came from the model's training data. Businesses should screen outputs that will be published or commercialized, particularly images, music, code and text that closely resemble known works, and avoid prompts that request material "in the style of" a specific living artist in ways that reproduce their expression. Questions about the legality of training models on copyrighted works are covered in our insight on AI training data and copyright risk.

AI-generated works in practice: sector examples

The practical impact of the originality test differs across industries. In advertising and marketing, agencies increasingly use generative tools for concept art, copy variations and product imagery. Brands should require agencies to disclose AI use, to keep source files and records of human edits, and to avoid delivering unaltered outputs as "original" campaign assets where exclusivity matters. Trademarks and registered designs may protect key brand elements more reliably than copyright in AI-generated works.

In software development, AI coding assistants generate significant volumes of code. Copyright in the overall program is likely to subsist where developers design the architecture and select, modify and integrate the generated code, but individual snippets may not be protected. Development teams should track AI-assisted code, review it for licence compliance and security, and rely on confidentiality and trade secret protection for proprietary systems.

In publishing, games and media, where licensing revenue depends on exclusive rights, producers should be cautious about building franchises on characters or worlds generated mainly by AI. Human-authored designs, scripts and edits give a firmer foundation for licensing and enforcement. See our insight on video game law.

Deepfakes, personality rights and compilations

Deepfakes raise issues beyond copyright. Using a person's likeness or voice without consent can engage provincial privacy statutes and torts, the common-law tort of misappropriation of personality, Quebec's Civil Code protection of reputation and privacy, defamation and, for intimate images, the Criminal Code. Advertising that uses a synthetic likeness of a real person may also be misleading under the Competition Act. Businesses producing synthetic media should obtain written consents and releases that expressly cover AI-generated likenesses and voices.

Compilations of AI-generated content, such as datasets, image libraries or anthologies, may be protected as compilations if the selection or arrangement reflects human skill and judgment, even where individual items are not protected. The protection is thin: it covers the selection or arrangement, not the underlying items. A competitor that independently assembles a different selection of similar AI-generated material would not infringe, so dataset owners usually rely on contracts, access controls and confidentiality as well.

An internal policy for generative AI use

Most ownership problems can be prevented with a short internal policy. A practical policy for creating AI-generated works would list approved tools and the terms under which they may be used; prohibit entering confidential information or personal information into tools that do not protect it; require human review and substantive editing of outputs intended for publication or commercial use; require records of prompts, iterations and edits for important works; set rules for using third-party names, likenesses and styles; and assign responsibility for checking outputs against known works before release. The policy should be aligned with privacy, procurement and information security rules, and reviewed as tools and the law change.

Policies are also persuasive evidence. If a business later needs to show that a work reflects human skill and judgment, a documented workflow requiring human creative control, followed in practice, supports that claim.

Policy outlook

The federal consultation on copyright in the age of generative AI (ISED) asked whether the Act should clarify authorship and ownership of AI-assisted works, whether text and data mining exceptions are needed, and how liability for infringing outputs should be allocated. Other jurisdictions have taken different approaches, and international developments will influence Canadian policy. Until Parliament acts, Canadian courts will apply the existing originality test. Businesses should plan on the assumption that only human-authored contributions are reliably protected, and revisit that assumption when courts or Parliament give clearer guidance. See our pages on copyright and trade secrets and artificial intelligence.

How TRW & Co helps

We advise creators, publishers, technology companies and brands on AI-generated works and copyright in Canada. We review AI tool terms, draft employment, contractor and licensing agreements that address AI-assisted work, design documentation practices that support authorship, assess infringement and personality rights risks in synthetic media, and help clients respond to disputes. To discuss your use of generative AI or a dispute over AI-assisted content, contact us. This article is general information, not legal advice.

Frequently asked questions

Is AI-generated art protected by copyright in Canada?

Only to the extent it reflects a human author's original expression. Under CCH Canadian Ltd. v. Law Society of Upper Canada, 2004 SCC 13, originality requires skill and judgment that is not purely mechanical. Output produced autonomously with minimal human input is unlikely to qualify; human selection, arrangement and editing may.

Can an AI system be an author under the Copyright Act?

The Act assumes human authors: subsistence depends on the author's nationality or residence, the term is measured by the author's life, and moral rights attach to authors. An AI system cannot hold rights, so the better view is that it cannot be an author, though courts have not yet decided the point.

Does registering an AI-assisted work prove copyright?

A certificate of registration is evidence that copyright subsists and that the registered person is the owner, but the presumption can be rebutted. The Copyright Office does not examine whether a work is original, so registration does not settle questions about AI authorship.

Who owns outputs from a generative AI tool?

Between the parties, ownership is governed by the tool's terms and by employment and contractor agreements. Contracts can allocate whatever rights exist but cannot create copyright where none subsists, so businesses should also rely on confidentiality, trademarks and other protections.

Has Canada changed its copyright law for AI?

Not as at the consolidation current to 19 October 2026. The federal government consulted on copyright and generative AI, with submissions closing on 15 January 2024, and published a report on the feedback, but no AI-specific amendments have been enacted.

Sources

  1. Copyright Act, R.S.C. 1985, c. C-42
  2. Supreme Court of Canada, docket 29320 (CCH Canadian Ltd. v. Law Society of Upper Canada)
  3. ISED, Consultation on Copyright in the Age of Generative Artificial Intelligence

General information only, not legal advice. References to statutes, courts and authorities describe published law and practice at the date shown.