Every ownership rule in the Copyright Act starts from the same word: author. Reading the Act’s own definitions and default rules closely is a more reliable way to reason about AI-generated material than reaching for an analogy from a different country’s law.
Key takeaways
Section 2 does not define “author” itself, which is part of why this stays a hard question — the Act uses the word constantly without ever pinning down exactly who or what can hold the role. What it does define, and define precisely, is what happens once authorship is established: who owns the resulting copyright, and how that ownership can move to someone else.
Section 13(1) states the general rule plainly: the author of a work is the first owner of the copyright in it. Everything else in the Act’s ownership scheme is an exception to that default, which is exactly why the default itself is worth reading carefully when AI is involved — if there is no author in the sense the Act contemplates, the default has nothing to attach to.
Section 13(3) is the most useful comparison point the Act actually contains. Where an employee creates a work in the course of their employment, the employer — not the employee — is the first owner, absent an agreement to the contrary. treadstonelaw.ca explains how this plays out for software specifically: an employer generally owns code its employees write on the job, without needing a separate assignment, because the Act already routes ownership to the employer by default. This shows the Act is entirely comfortable separating the person who did the work from the person who owns the result — but the mechanism that does the separating is an employment relationship between two legal persons, which is not the same relationship a business has with an AI tool it operates.
It is tempting to reach for the United Kingdom’s computer-generated-works provision, which names the person who made the arrangements necessary for the work’s creation as its author. Nothing like it exists in Canada’s Copyright Act. Treating Canadian law as though it had quietly adopted the same rule is a common and understandable shortcut, and it is not accurate — the honest position is that Parliament has simply not addressed the scenario, not that it addressed it the same way another country did.
The form itself makes the same assumption. CIPO’s own Application for Registration of a Copyright in a Work instructs, in its Author section, that “the individual who created the work should be named as author” — there is no field for naming a tool, only a person.
Whatever the answer on authorship turns out to be for a specific piece of AI-assisted work, Section 13(4) already governs how any interest in it moves between parties: no assignment or grant of a licence interest is valid unless it is in writing and signed by the owner. A business that commissions AI-assisted work from a vendor, and wants to be certain it owns the result, gets there the same way it always has — a written agreement that says so — rather than through any special AI-specific mechanism, because no such mechanism exists.
Buyers in Canadian AI-business sales are already asking who owns a model trained on licensed data, and the underlying reason is exactly this authorship gap: a buyer cannot value what it cannot confirm someone owns. A business that can show, work by work, whether a piece was human-authored, AI-assisted with substantial human input, or essentially raw model output — and who if anyone holds rights in each category — is answering the question a diligence process is actually built to ask, well before anyone raises it.
It helps to separate three things the Act keeps distinct, because AI-generated material tends to blur them together in conversation. Authorship is the question of who created the work, and the Act never quite defines it. Ownership is a separate question the Act answers precisely once authorship is settled — the author owns it, unless employment or assignment says otherwise. Licensing is a third, entirely mechanical question of how an owner grants use to someone else, governed by the writing-and-signature requirement regardless of what the underlying work is. A business unsure about AI-generated material is usually actually unsure about the first question; the second and third already have clear answers once the first is resolved one way or the other.
A marketing team asks a copywriting tool to draft a product description, then a staff writer substantially rewrites it, restructures the argument, and adds original detail the tool never produced. Under Sections 13(1) and 13(3), the finished piece looks like ordinary employee-authored work: the writer did real authorial work, did it in the course of employment, and the employer owns the result by the Act’s normal default — the same outcome as if the writer had started from a blank page instead of a draft. The analysis only gets harder as the human contribution shrinks toward nothing, which is exactly the scenario the Act was never written to resolve.
No. The Act uses the term throughout without a specific definition, which is part of why AI-generated material sits in genuinely open legal territory rather than a clearly resolved one.
The Act's default rule speaks in terms of an author who is a person, with ownership then flowing to an employer in the employment scenario — that is a distinct question from authorship itself, which the Act still frames around a human creator.
Not automatically, and not by any specific rule in the Act. Paying for a tool's subscription is not the same legal event as authorship or a written assignment, which is why a business relying on AI-generated material benefits from documenting the human creative input involved, or securing rights contractually wherever it can.
Not for the authorship analysis itself — the Copyright Act does not distinguish between one generative tool and another. What can matter is what that specific tool's own terms of service say about the rights it grants a user in the output, which is a contractual question layered on top of, not a substitute for, the copyright question.
It is a reasonable practice given how much the authorship analysis above turns on the extent of human involvement. Drafts, edit history and a note of who made substantive creative decisions are the kind of evidence that would actually matter if authorship or ownership of a specific piece were ever challenged.
Related: whether AI output is protected by copyright at all and who owns AI-generated work in Canada.
A due-diligence process is where an unresolved authorship question turns into a real number on a term sheet.
