Canada's Strong and Free Elections Act — Bill C-25 — received royal assent on June 18, 2026, giving Ottawa its first criminal-law answer to AI-generated deepfakes of candidates and election officials. Six weeks later, the law is drawing a specific and credible critique: the intent standard that makes it constitutionally defensible may also make it functionally unenforceable.
What the law actually does
Bill C-25 amends the Canada Elections Act to add a new offence, in the amended s.480.1, covering the creation or distribution of an image or voice imitation "likely to be mistaken for" a candidate, party leader, or the Chief Electoral Officer, where the imitation is made with intent to mislead voters. Convictions are dual-procedure: up to 14 years' imprisonment if prosecuted on indictment, or a $2,000 fine and up to three months in jail on summary conviction, per s.500(5) as amended. Parliament also raised the ceiling on administrative monetary penalties the Commissioner of Canada Elections can levy for related violations, from $1,500 to $25,000 for individuals and from $5,000 to $100,000 for organizations (Canada.ca, June 2026; Parliament of Canada, Bill C-25 third reading). The act carves out an explicit defence for parody and satire.
The steelman
The case for legislating now is genuinely strong. Generative AI has collapsed the cost of producing a convincing fake video or voice clone of a sitting minister or a party leader from studio budgets to a laptop and an afternoon. Canada's previous impersonation rules, dating to a pre-generative-AI Elections Act, covered someone falsely claiming to be a candidate — not a synthetic likeness that never makes that claim but is indistinguishable from the real thing. That was a real gap, and other democracies (the EU's AI Act transparency rules, several U.S. state deepfake statutes) have moved on the same problem. A criminal offence with a 14-year ceiling signals, credibly, that Ottawa treats synthetic election manipulation as a serious harm rather than a labeling nuisance. Given founded anxiety about foreign interference in Canadian elections, building a deterrent into the Criminal Code rather than leaving it to platform terms of service is a defensible legislative choice.
Why the intent bar is the whole story
But a criminal offence needs a mens rea the Crown can prove beyond reasonable doubt, and "intent to mislead" is precisely the element that is hardest to establish for AI-generated content. As the Hill Times reported on July 13, 2026, Bloc Québécois House Leader Christine Normandin said Parliament is "kind of walking in the dark" on how the provision will actually function, and legal observers quoted in the same piece were blunter still about the intent element: "I'm not seeing how it can be done" (Hill Times, July 13, 2026).
The practical problem is threefold. First, AI tools generate content probabilistically from prompts a user may not have written with any electoral figure in mind — proving that a specific creator subjectively intended to mislead voters, rather than experimented, satirized, or was careless, is a different evidentiary task than proving they clicked "post." Second, criminal prosecutions in Canada routinely take months to years to reach trial; a deepfake timed for the final week of a 37-day writ period will have done its damage — and likely been forgotten — long before any charge is laid, let alone proven. Third, the law's own parody and satire defence, necessary to survive Charter scrutiny under s.2(b), gives any defendant a ready-made argument that blurs the line prosecutors must draw.
The platform gap
The topic this law was billed as addressing — misinformation spreading on platforms during a campaign — is only partially in its scope. Bill C-25 does not impose a direct takedown obligation on platforms the way the EU's Digital Services Act does for illegal content; instead, the Commissioner of Canada Elections may request removal of offending content, a request platforms can decline absent a court order. That leaves the fastest-moving vector for election deepfakes — viral spread on X, TikTok, or WhatsApp within hours of posting — governed by voluntary platform policy, not statute, even as the criminal provisions target the much slower question of who made the content and why.
Our view
The intent requirement is not a drafting error to be fixed in the next amendment — it is the constitutionally necessary price of writing a criminal deepfake law that does not also criminalize satire, commentary, and legitimate AI-assisted campaign content. A lower bar (strict liability for any misleading synthetic media of a candidate) would chill exactly the kind of parody and rapid-response content that campaigns and comedians alike produce constantly, and would hand incumbents a tool to threaten critics with prosecution. Canada was right to build in that friction. The honest conclusion, though, is that C-25's criminal provisions will rarely if ever be the thing that stops a deepfake during an actual campaign — that job falls to platforms, fact-checkers, and Elections Canada's public-facing rapid-response capacity, none of which this bill meaningfully strengthens. Parliament should say so plainly rather than let a law with a 14-year ceiling imply a deterrent capacity it cannot deliver in the 37 days that actually matter.