Deepfake Recordings Allegedly Influence Slovakian Election

Days before Slovakia's election, deepfake audio recordings surfaced, allegedly featuring conversations between a journalist and a leading liberal politician discussing vote-rigging and other controversial topics. The recordings were spread on social media platforms and may have influenced the election outcome, which saw the pro-Russian populist party winning.

Risk classification

• Primary risk domain: 4 Malicious actors
• Primary risk subdomain: 4.1 Disinformation, surveillance, and influence at scale

The incident involved coordinated, at-scale dissemination of AI-generated deepfakes targeting the Slovakian electorate to manipulate public opinion and influence the outcome of a national election.

Additional risk subdomains

• 3.1 False or misleading information: The AI system was used to generate highly realistic but entirely fabricated audio recordings of political figures and journalists to mislead voters.
• 4.3 Fraud, scams, and targeted manipulation: The deepfake involving Todova was used to target, smear, and discredit a specific prominent journalist and media outlet.

Causal factors

• Entity: Human
• Intent: Intentional
• Timing: Post-deployment

The incident was driven by human actors intentionally using deployed AI voice-cloning tools to generate and spread deceptive political disinformation during an election campaign.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The incident involves AI-generated content, specifically deepfakes, which under the EU AI Act are subject to transparency obligations to ensure users are informed of their synthetic nature.

AI system and alleged parties

• AI system: None named
• AI purpose: Voice Generation; Text Style Replication
• Behaviour type: Tool
• Alleged developer: unknown
• Alleged deployer: unknown
• Alleged harmed parties: Slovakian electorate, Monika Todova, Michal Šimečka, Democratic process in Slovakia

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Minor. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Minor, indirect Minor
• Differential treatment: direct Negligible, indirect Minor
• Civil rights: direct Minor, indirect Minor
• Democracy: direct Minor, indirect Minor
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Minor, indirect Negligible
• Epistemic: direct Minor, indirect Minor
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

Reported: Yes, the reports describe the creation and spread of malicious deepfakes and hate speech.

Directly caused: AI-generated voice clones were used to fabricate conversations about election rigging and beer prices, which spread to hundreds of thousands of users on Facebook, Instagram, and Telegram.

Indirectly caused: The far-right party Republika used AI-generated voices in ads that also attacked the LGBTI+ community, spreading transphobic hate speech.

Differential treatment

Reported: Yes, the reports describe attacks on the LGBTI+ community.

Indirectly caused: The far-right party Republika used AI-generated campaign videos to attack the LGBTI+ community, promoting discriminatory narratives.

Civil rights

Reported: Yes, the reports describe attempts to interfere with democratic voting rights and press freedom.

Directly caused: The deepfakes violated the civil rights of political candidates and journalists by fabricating statements to manipulate an election.

Indirectly caused: The dissemination of deepfakes during a pre-election moratorium undermined the electorate's right to receive accurate information to make an informed democratic choice.

Democracy

Reported: Yes, the reports explicitly describe the erosion of democratic norms through election interference.

Directly caused: The deepfakes were strategically released during a 48-hour pre-election moratorium to manipulate voter sentiment and disrupt the integrity of the Slovakian parliamentary election.

Indirectly caused: The incident demonstrated how AI-generated disinformation can exploit regulatory gaps and platform moderation failures to influence democratic processes.

Inferred additional harm: The success of such campaigns may encourage future domestic and foreign actors to deploy similar AI-driven tactics to subvert democratic elections globally.

Psychological

Reported: Yes, the reports describe distress and reputational smearing of targeted individuals.

Directly caused: Monika Todova expressed being 'disgusted' by the deepfake manipulating her voice to fabricate a corrupt conversation.

Inferred additional harm: The targeted political figures and journalists likely experienced significant stress, anxiety, and reputational distress due to the viral spread of highly realistic fabrications.

Epistemic

Reported: Yes, the reports describe the spread of fabricated audio recordings that misled the public.

Directly caused: AI voice cloning was used to fabricate a two-minute conversation about vote-rigging between a politician and a journalist, which was widely believed by some segments of the public.

Indirectly caused: The incident contributed to the erosion of shared truth, making it harder for citizens to distinguish between authentic and synthesized recordings of public figures.

Inferred additional harm: The proliferation of such deepfakes may lead to a 'liar's dividend,' where public figures can falsely claim authentic but damaging recordings of themselves are AI-generated fabrications.

People affected

• Occurrences reported: 1
• People reportedly harmed: 3
• People reportedly exposed: 300000

Potential causes

Management

• Inadequate Platform Safety Spending: Platforms failed to invest sufficiently in content moderation for small nations.
• Delayed Government Response: Slovak authorities failed to proactively prepare for AI-driven hybrid threats.

Technology

• Easy Access to Voice Cloning Tools: Apps like HeyGen made creating realistic voice clones cheap and accessible.
• Inadequate AI Detection Tools: Fact-checkers lacked reliable, accessible tools to quickly identify deepfakes.

Data Inputs

• Abundant Public Audio Samples: Public appearances of politicians provided ample data to train voice clones.

Human Factors

• Voter Susceptibility to Hoaxes: Many voters believed the fakes because they aligned with existing biases.
• Politicians Spreading Disinformation: Slovak politicians actively shared deepfakes to discredit opponents.

Process and Methods

• Exploitation of Election Moratorium: Releasing fakes during the 48-hour moratorium prevented timely debunking.
• Policy Loophole for Audio Deepfakes: Meta's manipulated-media policy initially only covered fake videos, not audio.
• Inconsistent Platform Moderation: Social media platforms responded slowly and inconsistently to reported fakes.

Regulatory Environment

• Lack of Effective AI Regulation: No robust laws existed to prevent or penalize election-disrupting deepfakes.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide consistent, detailed accounts of the deepfake incidents, the specific individuals targeted, the platforms used for dissemination, and the fact-checking efforts that confirmed the use of AI voice cloning. The role of the AI systems is explicitly stated, and multiple independent sources (AFP, Reset, Demagog) corroborate the events.
• Ambiguities identified: The exact identity of the creators of the deepfakes remains unconfirmed, though there are strong suspicions of Russian state involvement or pro-Russian domestic actors.

AI-generated deepfake audio was deployed in Slovakia to influence a national parliamentary election during a critical 48-hour pre-election moratorium. This incident represents a substantial threat to sovereignty and information security, establishing a major precedent for AI-driven election interference in democratic nations.

• Overall national security impact: Substantial
• Scope: Single nation
• Primary target: Slovakia
• Alleged perpetrator: Unknown

Threat characteristics

• Imminence: Immediate. The attack was strategically timed during a 48-hour pre-election moratorium, requiring immediate, real-time crisis response that overwhelmed existing moderation capabilities.
• Autonomy: Human-controlled. The AI voice-cloning systems functioned strictly as tools, requiring direct human input of text and voice samples to generate the deceptive audio.
• Novelty: First-of-its-kind. Widely considered one of the first major instances of AI-generated deepfake audio being deployed to influence a national election within the European Union.

Impact by dimension

• Physical security: Negligible. No physical targets, kinetic systems, or critical infrastructure were affected or targeted in this incident.
• Information security: Substantial. Sophisticated AI-generated voice clones were used to conduct a targeted disinformation campaign against a political leader and journalist, exploiting social media to reach hundreds of thousands of voters.
• Sovereignty: Substantial. The deepfakes directly targeted a national parliamentary election during a sensitive 48-hour pre-election moratorium, aiming to manipulate democratic processes and voter sentiment.
• Economic security: Negligible. No strategic technology theft, financial system attacks, or significant economic impacts were reported.
• Societal stability: Minor. The incident eroded epistemic trust and targeted specific public figures, representing a minor threat to societal cohesion and democratic norms, but did not cause civil unrest.