An AI-generated audio clip put words in the mouths of a party leader and a journalist, surfaced inside the pre-election silence period when neither could properly respond, and offered an early preview of how cheap synthetic media can be timed to slip past a democracy’s defences.
In the last days of September 2023, as Slovakia headed toward a parliamentary election, an audio recording began spreading on Facebook. In it, a voice that sounded like Michal Šimečka, leader of the liberal Progressive Slovakia party, appeared to discuss rigging the election by buying votes from the country’s Roma minority. His apparent interlocutor was a voice resembling Monika Tódová, a well-known journalist at the newspaper Denník N. A second clip had the same Šimečka-like voice joking about doubling the price of beer. Both recordings were fabrications, generated with artificial intelligence. Both were released at the worst possible moment for anyone trying to correct them.
The Crucial Timing
Slovakia observes a moratorium in the final 48 hours before polls open, a quiet period during which media and politicians are restricted from campaigning. The fake audio surfaced squarely inside that window. As the International Press Institute noted, the moratorium left both the falsely quoted politician and the falsely quoted journalist in an almost impossible position. Šimečka and Tódová denounced the clips as fakes, but the rules that are meant to keep the final hours of a campaign calm also muffled the very people best placed to knock the story down.
There was a second gap. The recordings were audio, not video, and at the time the platforms’ policies on manipulated media were written mainly around video. That mismatch, examined in Reset’s analysis of the case, meant the clips did not fit neatly into the categories moderators were equipped to act on quickly. A fabrication landed in the seam between the election law and the platform rules, and it sat there through the hours that mattered most.
Why audio, and why these two people
Synthetic audio is easier to fake convincingly than video, because the ear forgives imperfections that the eye would catch, and because a phone recording of a private conversation is expected to sound rough. A grainy, low-quality clip of two people talking is exactly the kind of artefact that carries no visual tells and invites the listener to fill in the gaps.
The choice of targets was equally deliberate. Pairing a party leader with a respected journalist manufactured a story about collusion between politics and the press, one of the most reliably corrosive narratives available, and it did so days before voters made their decision. The security researcher Bruce Schneier, writing about the episode shortly afterward, flagged it as one of the first cases in which a deepfake was deployed to influence a national election at the decisive moment.
Attribution, honestly
It is important to be careful about who did this. The clips carried pro-Kremlin fingerprints in their framing and served a pro-Russian outcome, and the affair has been widely discussed as an example of the disinformation playbook Russia has used across the region. But the operation was not definitively traced to the Russian state, and researchers who studied it, including the team behind a detailed review published by Harvard’s Misinformation Review, cautioned against both overstating the deepfake’s decisive impact and overstating the certainty of attribution. What can be said is that the method fit the pattern of pro-Russian information operations, and that the beneficiary of a weakened liberal party was clear enough.
Outcome and what changed
Robert Fico’s Smer party won the election and returned to power on a platform sceptical of arming Ukraine, and it is tempting to draw a straight line from the fake audio to the result. The honest position is that no one can measure how many minds the clips changed, and the Harvard review pushed back hard on the easy headline that a deepfake had swung the vote. The danger the clips demonstrated was not their impact on that one election but their timing, cheapness, and precision.
Since then, the Slovak case has become a fixed reference point for everyone thinking about elections and synthetic media, cited across Europe as lawmakers debated deepfake-labelling rules and platforms rewrote manipulated-media policies to cover audio as well as video. The clip itself was disposable and is long forgotten. The vulnerability it exposed is not, because every democracy with a pre-election silence period still faces the same trap: the people best equipped to correct a fabrication may be the ones the rules silence, and the correction may arrive only after the ballots are cast.
Leave a Reply