Should AI-Generated Political Deepfakes Be Criminalised?

Asked as a flat question, this sounds like a referendum on a technology, but the serious version is narrower. The issue is not whether AI-generated media should be banned, but whether the law should criminalise deepfakes deployed to deceive voters during an election campaign. I will argue that it should, through a tightly drawn offence that requires an intent to mislead voters, is confined to the regulated campaign period, and carves out satire, journalism and art. A blanket ban on the technology would be a different and indefensible thing.

The case for treating these as a distinct wrong rests on what a deepfake actually does. An ordinary political lie still has to persuade; it competes in the marketplace of ideas, where more speech can answer it. A fabricated video of a candidate does not argue, it impersonates, hijacking the viewer's senses before any reasoning begins. The clearest illustration is the 2024 New Hampshire robocall, which cloned President Biden's voice to tell Democrats to stay home in the primary. It did not make a false claim to be rebutted; it counterfeited the man himself. When deception operates below the level of deliberation, the comforting answer that counterspeech will cure it simply runs out of time.

The harm, moreover, falls on a public good rather than a private reputation. Defamation protects an individual's standing; a deepfake released days before a poll injures the electorate's capacity to choose, and the 'liar's dividend' compounds the damage, because once fabrication is common, genuine recordings can be dismissed as fakes too. Canada has already recognised this institutional harm. Bill C-65, introduced in 2024, extended the impersonation and misleading-publication offences in the Canada Elections Act so that they apply regardless of the medium, expressly to capture deepfakes that make candidates appear to say what they never said, while leaving satire and parody untouched. Civil remedies cannot do this work, because they are structurally too slow: the Zelensky 'surrender' deepfake of March 2022 was removed by the platforms within hours, yet it had already spread, and worked at all only because Ukraine had pre-warned the public. Most deepfakes will not be pre-empted that way, and an injunction obtained after the count protects no one.

The strongest objection is free expression, and it must be taken seriously. Article 10 of the Convention protects even false statements in principle, and in United States v Alvarez (2012) the US Supreme Court struck down a law criminalising lies about military honours, holding that falsity alone does not forfeit protection. A clumsy offence against 'deceptive synthetic media' would indeed sweep in comedy impressions and obvious caricature, punishing the lie rather than the harm. But Article 10 is a qualified right: restrictions are permitted where they are prescribed by law, pursue a legitimate aim and are necessary in a democratic society, and protecting the integrity of elections is a paradigm legitimate aim. The answer to Alvarez is not to abandon the offence but to draft it precisely, keyed to an intent to deceive voters, limited to the campaign window, with satire and journalism expressly carved out.

Two further worries push toward modesty rather than abandonment. First, existing law may already reach the worst cases: the Biden robocall drew a six-million-dollar FCC forfeiture under the Truth in Caller ID Act and criminal charges for voter suppression and candidate impersonation, with no bespoke deepfake statute in sight. A new offence must therefore add precision, not duplication. Second, the harm is fast, anonymous and cross-border, while prosecution is slow, so criminal law is necessary but not sufficient. This is why the offence should sit above a lighter-touch layer: the EU AI Act's duty to disclose AI-generated audiovisual content targets the real wrong, undisclosed deception, while leaving labelled satire alone.

The defensible conclusion, then, is a regulated yes. Criminalise the intentional, campaign-period deepfake that is designed to deceive voters about a candidate or the vote, carve out parody, satire, art and journalism, and underpin the offence with a mandatory disclosure regime. Reject, equally firmly, any blanket prohibition on synthetic media as such. The law's task here is the old one in new dress: to separate liberty from licence, protecting the comedian and the journalist while denying the forger the power to counterfeit a democracy.