Skip to content
Influence

Deepfakes in elections

Intermediate9 min read

A deepfake is synthetic audio, image, or video generated by AI to make someone appear to say or do something they never did. In elections the danger is twofold: a convincing fake released too late to debunk, and the "liar's dividend" — real evidence waved away as fake precisely because deepfakes now exist.

In this article

What a deepfake actually is

A deepfake is media — audio, image, or video — synthesised by AI so that a real person appears to say or do something they never did. The technology is not new, but two things changed recently: it became cheap, and it became good enough that a casual viewer cannot tell. In an election, that combination turns a person's own voice and face into something an adversary can borrow.

The popular image of the threat is a flawless fake video of a candidate. The reality so far is more modest and more clever: the most effective political deepfakes have been audio, not video — because audio is cheaper to generate, carries none of the visual tells that expose fake video, and travels invisibly through voice notes in private messaging apps.

The Slovakia case: a fake in the silence

The reference example is Slovakia's parliamentary election of 30 September 2023. Roughly 48 hours before polls opened, an audio clip spread on social media purporting to capture Michal Šimečka, leader of the liberal Progressive Slovakia party, and a journalist from the newspaper Denník N discussing how to rig the vote by buying ballots from the Roma minority. The conversation never happened; the audio was a fabrication.

What made it dangerous was not the quality of the fake but the timing. It landed inside Slovakia's 48-hour pre-election moratorium — a legally mandated silence period when media and politicians are restricted from campaigning — so the very people best placed to debunk it were constrained from responding loudly while it circulated. Progressive Slovakia went on to lose the election narrowly.

Here the "0 bullshit" standard matters. It is tempting to call this "the deepfake that decided an election." Careful academic analysis of the Slovak case found the opposite kind of certainty is unwarranted: the fake's actual effect on the result is impossible to isolate from everything else that moved votes. The honest lesson is about the method — a fake weaponised by timing — not a proven change of outcome.

The two threats: the fake and the liar's dividend

Deepfakes threaten elections in two distinct ways, and the second is easy to miss.

  1. The fabricated clip. A synthetic audio or video, released to smear a candidate or fake a scandal — ideally too late to rebut, as in Slovakia.
  2. The liar's dividend. The subtler harm. Once everyone knows convincing fakes exist, a politician caught on a real recording can simply claim it is a deepfake. The mere existence of the technology gives liars a ready-made escape hatch — so deepfakes corrode trust even when no fake is actually deployed.

The second is arguably the larger long-term danger. A single fake can be debunked; a general collapse of "seeing is believing" cannot be, and it damages the honest and the dishonest alike.

Why audio is the frontier

It is worth being concrete about why audio leads:

  • Cost. A usable voice clone needs only a few minutes of a target's real speech — abundant for any public figure.
  • No visual tells. Fake video still betrays itself in hands, teeth, blinking, and lighting. Audio has no equivalent giveaways for a casual listener.
  • Low fidelity hides flaws. A deliberately phone-quality clip makes any artefacts sound like a bad connection, not a fake.
  • Dark spread. Voice notes move through closed channels — WhatsApp, Telegram, Signal — where no fact-checker or platform can see them to intervene.

What actually protects an election

Detection tools help but always lag the fakes, so the durable defences are procedural and human:

  • Prebunking. Warning audiences in advance that last-minute fakes are likely, so a late clip is met with suspicion rather than belief.
  • Provenance over content. Judging a clip by where it came from — who posted it first, when, with what source — rather than by how convincing it sounds.
  • Fixing the silence-period trap. The Slovakia case exposed how moratorium rules can protect a fake; several jurisdictions are now rethinking how debunking is allowed during the quiet period.
  • Rapid, trusted rebuttal. A credible, fast denial from the impersonated person and independent outlets, ready to move the moment a fake appears.

None of this requires spotting the fake by eye — which is lucky, because that is getting harder every year. The practical skills for scrutinising a suspicious clip yourself belong to the Verify track, but the mindset starts here: when a shocking recording appears right before a vote, the first question is not "is this real?" but "where did this come from, and why now?"

Frequently asked questions

Have deepfakes ever changed an election result?

There is no clear, documented case of a deepfake decisively swinging a national result. The Slovakia 2023 audio fake is the most cited example, but careful analysis found its actual effect impossible to isolate. The realistic threat is corrosion of trust and last-minute confusion, not a proven change of outcome.

Why is deepfake audio more dangerous than video?

Audio is cheaper and faster to generate, lacks the visual glitches that give away fake video, and spreads easily through voice notes in private messaging apps where fact-checkers cannot see it. A phone-quality voice clip is also inherently low-fidelity, so imperfections read as a bad recording rather than a fake.

What is the liar's dividend?

It is the second-order harm of deepfakes: once people know convincing fakes exist, anyone caught on a real recording can claim it is a deepfake. The mere possibility of fakery lets bad actors dodge genuine evidence, so deepfakes damage trust even when no fake is actually made.

How can I tell if an election clip is a deepfake?

Slow down and check provenance before content: who first posted it, when, and is any original source or context attached? Be especially wary of emotionally charged clips that surface right before a vote, in low quality, with no clear origin. Detailed technique is covered in the Verify track.

Sources

  1. 1.Beyond the deepfake hype: AI, democracy, and 'the Slovak case'Harvard Kennedy School Misinformation Review (2024)
  2. 2.Slovakia: Deepfake audio of Denník N journalist offers worrying example of AI abuseInternational Press Institute (IPI) (2023)
  3. 3.1st EEAS Report on Foreign Information Manipulation and Interference ThreatsEuropean External Action Service (EEAS) (2023)
Written by VoteRights EditorialReviewed by VoteRights standards deskLast reviewed

Continue learning

Related cases in the database