Canada has no law that uses the word “deepfake.” What exists is a set of ordinary laws — on deceptive marketing, on personal information, on court evidence — that apply once a synthetic video, image or voice is used to do something a regulator or court already has a name for.
Key takeaways
The Competition Bureau’s own discussion paper on AI and competition is candid that it is exploring the issue, not settling it: it states plainly that the paper “is not meant to predict any outcomes or provide any recommendations”. (competition-bureau.canada.ca) That same paper is the closest thing to an official Canadian naming of the risk: it observes that generative AI “eases the ability to scale” deceptive marketing conduct, and that the technology “could be leveraged to make deceptive marketing conduct more convincing, using ‘deepfake’ (digitally altered) voices, images, or videos”. Naming a risk in a discussion paper is not the same as regulating it — no enforcement action or binding rule has followed from that specific paragraph.
Section 74.01 of the Competition Act does not need to mention AI to reach an AI-made claim. Paragraph (1)(a) reaches any representation to the public that is “false or misleading in a material respect”, and paragraph (1)(b) goes further for performance claims specifically: a representation “in the form of a statement, warranty or guarantee of the performance, efficacy or length of life of a product that is not based on an adequate and proper test thereof, the proof of which lies on the person making the representation”. (laws-lois.justice.gc.ca) Applied to a deepfake: if a business uses a synthetic video of a customer or celebrity endorsing a product to make a performance claim more convincing, the business making that claim carries the burden of having tested what it is claiming — the technology used to dress up the claim does not shift that burden anywhere else.
PIPEDA defines personal information simply: “information about an identifiable individual.” (laws-lois.justice.gc.ca) A recognizable face or voice used to build or run a deepfake is exactly that, which means a business collecting, using or disclosing it needs to meet PIPEDA’s ordinary rules — identifying the purpose, limiting use to that purpose, and having appropriate consent. Canada’s privacy commissioners’ own joint principles for generative AI add a further layer for the business deploying the system, not just the one whose likeness is used: they name “no-go zones” that include “the collection, use, or disclosure of personal information for purposes that are known or likely to cause significant harm to individuals or groups, or activities which are known or likely to threaten fundamental rights and freedoms”. (priv.gc.ca) A non-consensual deepfake built from someone’s personal information sits squarely inside that description.
A different tool applies where the harm is commercial confusion rather than personal misuse: passing off. Treadstone Law’s explainer describes it as “a common-law tort that protects the goodwill a business has built in its name, branding, or overall look, even without any trademark registration”, requiring proof of goodwill, a likelihood the public will be confused into thinking there is a connection between two businesses, and resulting or likely damage. (treadstonelaw.ca) That page addresses ordinary brand-goodwill disputes, not AI specifically — but its three elements apply just as well to a deepfake video impersonating a company or its spokesperson to create a false impression of endorsement: if a viewer is likely to be confused into believing the real business said or did something it did not, the same test decides whether that confusion is actionable.
Litigation is where Canadian institutions have moved fastest, because a court can regulate what happens in its own room without waiting for Parliament. The Federal Court’s notice on AI in filings gives the term its own working definition, footnoted for the purpose of that notice only: a “deepfake” is “AI-generated images of human subjects that either replace one person’s likeness convincingly with that of another, or that do not exist in real life”. (fct-cf.ca, Notice to the Parties and the Profession, May 7, 2024) That definition exists to decide when a party must declare that AI produced content in a court filing — it is not a general-purpose legal definition and has no force outside that specific notice. Beyond it, no Canadian court decision on an unauthorized deepfake is cited here: this environment cannot reach CanLII, so this piece names the applicable statutes and doctrines rather than a case that decided them.
A regional retailer discovers a video circulating online that uses AI to make it look like its (real, named) store manager is offering a discount code that does not exist. Three separate legal threads apply at once, for three different reasons: the false discount claim is reviewable conduct under the Competition Act regardless of who made the video, because it is a misleading representation to the public; using the manager’s real likeness without consent to make the video is a collection and use of personal information under PIPEDA if any business is behind it; and the retailer’s own passing-off claim turns on whether viewers are likely confused into thinking the retailer authorized the offer — a real prospect given the video uses its actual employee and branding. None of the three requires a “deepfake law” to exist.
Related: the same analysis applied to a cloned voice, why proving a video is synthetic is a separate problem from regulating it, how an existing duty of confidence constrains AI use generally
It depends entirely on what it is used for. There is no dedicated statute, but a commercial deepfake can trigger the Competition Act’s rules on false and misleading representations, a deepfake built from someone’s face or voice can trigger PIPEDA if a business is involved, and a deepfake that creates false brand confusion can support a passing-off claim.
Under the Federal Court’s own definition, used only for its litigation-disclosure notice, no — the notice states it “does not apply to AI that lacks the creative ability to generate new content,” and gives an example: it “does not apply to AI that only follows pre-set instructions, including programs such as system automation, voice recognition, or document editing.” (fct-cf.ca, Notice, May 7, 2024) Ordinary editing tools fall outside that specific definition.
That is likely both a Competition Act problem, because a false representation was made to the public, and a passing-off problem, because customers may be confused into thinking the company authorized it. The two claims run on different tests and can be pursued together.
Checking whether media, claims or a target’s own AI use are what they appear to be is a recurring diligence task, not a one-off legal question.
