“Who owns this” sounds like one question. For AI-generated work in Canada, it is actually three layered questions, and getting the answer right means working through them in order rather than guessing at the end result.
Key takeaways
Work through this in the order a court would rather than jumping straight to who paid for the tool. First: was there enough human authorial judgment — selecting, editing, arranging — for the result to be a work with an author at all, a question a companion piece on this hub covers in full. If the answer is no, none of the ownership rules below have anything to attach to, and the material likely sits outside copyright protection altogether.
Section 13 is unambiguous: the author of a work is its first owner, unless the work was made by an employee in the course of employment, in which case the employer owns it absent an agreement otherwise. Applied to AI-assisted work, this means a staff member who does the substantive human authorship — directing, editing, arranging AI output into a finished piece — produces something the employer owns by the ordinary default, the same as any other work product the employee creates on the job.
The employment default does not extend to independent contractors, who generally keep ownership of what they create absent a written assignment — even where they were paid for the work. A freelance designer or copywriter using AI tools as part of their own process still owns the finished piece by default, exactly as they would if they had used no AI tool at all. A business that wants to own AI-assisted work its contractors produce needs the same thing it would need for any other contractor work: a written assignment, not an assumption that payment settles the question.
Where the AI tool itself is a third-party product, its own terms of service sit on top of everything above, and a companion piece on this hub is the place this gets worked through in detail: what the vendor actually grants, whether that grant is an assignment or a licence, and the limits of what a vendor can grant at all. A business can have a perfectly clean answer on the employment or contractor layer and still find the vendor’s terms are the layer that actually determines what it can do with the result — reuse it commercially, sublicense it, exclude the vendor from any further claim.
A separate wrinkle appears where the AI-assisted work draws on material the business itself licensed in from somewhere else — stock photography, a licensed dataset, a contributor’s article. The same written-assignment formality already applies to that underlying material regardless of whether AI touched it afterward: a business can only pass along the rights its own upstream licence actually grants, and running licensed-in material through an AI tool does not launder or expand the scope of that original licence.
treadstonelaw.ca’s explainer on employer ownership of employee-written software is the clearest real-world illustration of how the Act separates the person doing the work from the person owning the result: an employer generally owns software its employees write on the job without needing a special assignment, because the employment relationship itself does the work. AI-assisted output produced by an employee sits on the same footing, provided the human authorship threshold above is met — the tool used to produce a first draft does not change which employment-law default applies to what the employee then does with it.
A marketing coordinator, an employee, uses a paid AI tool to draft and substantially edit a client-facing report. Layer one: the coordinator’s editing and arranging is real authorial work, so ordinary copyright analysis applies. Layer two: the coordinator did this in the course of employment, so under Section 13(3) the employer owns the result, not the coordinator personally. Layer three: the AI vendor’s terms assign output rights to the account holder, which here is the employer’s own corporate account — so the vendor layer agrees with, rather than overrides, the employment-law answer. Change any one fact — a contractor instead of an employee, a personal account instead of a corporate one — and the answer at the end can change too.
Change one fact from the example above: the marketing coordinator is not an employee but a freelance contractor, engaged with no written agreement beyond an invoice for hours billed. Layer one is unchanged — the contractor did real authorial work. Layer two now points the other way: absent a written assignment, the contractor, not the business that paid the invoice, owns the copyright in the finished report. Layer three may point either way depending on whose AI account was used and what that vendor’s terms say about whose account holds the output rights. A business relying on invoiced freelance work believing it automatically owns the result is relying on layer two working the way it does for employees, when the default for a contractor runs in the opposite direction.
None of this requires a new legal instrument invented for AI. The gap between what a business assumes it owns and what it actually owns closes the same way it always has for contractor work generally: a written agreement, signed before or at the point of engagement, that explicitly assigns the finished work — AI-assisted or not — to the business. The mistake worth avoiding is treating “we paid for it” or “we prompted the AI tool” as doing the work a signature is required to do.
By default, the freelancer, unless a written agreement assigns the rights to the business that hired them — the same rule that applies to any contractor work, AI-assisted or not.
It can, because some vendors' terms tie the output-rights grant to the account holder specifically. A business relying on employees' personal AI accounts for work product should check whether that changes who the vendor considers the rights-holder.
No single rule covers every case, which is exactly why this works as a layered analysis — the authorship threshold, the Copyright Act's ownership defaults, any written agreement, and the AI vendor's own terms all have to be checked, in that order.
Related: what the Copyright Act says about authorship and what AI vendor terms say about output ownership.
A custom engagement is the moment to put the employment, contractor and vendor layers in writing, rather than sort them out after the fact.
