Slovakia - Deepfake Audio Before Elections

What Happened

In September 2023, just 48 hours before Slovakia's parliamentary elections—during a strict, legally mandated media silence period—a highly realistic deepfake audio recording was published on Facebook. The cloned audio depicted Michal Simecka, leader of the pro-European Progressive Slovakia party, and a prominent journalist discussing how to rig the election, including buying votes from marginalized groups. Because of the media silence, mainstream journalists and candidates were legally restricted from reporting on or debunking the recording. The deepfake spread rapidly on social media, influencing public perception before being identified as a fake, contributing to the election victory of his pro-Russian rival party.

How MEG 1 would have acted

The deepfake represents a hostile, coordinated cognitive manipulation attack designed to bypass legal safeguards and subvert a democratic election, directly violating the cognitive sovereignty of the Slovak electorate.

MEG 2 Analysis - Legal Framework

Agent level: N3 (individualized/advanced) - generative audio-cloning AI system capable of real-time, highly realistic biometric simulation used to execute political and democratic manipulation.
Cause (6.1):
(c) hostile weaponization / biometric impersonation - the deliberate creation and deployment of cloned biometric voice assets designed to deceive a public electorate and disrupt a democratic election.
Attachment of liability: The unknown actors who created and distributed the deepfake are fully liable; the social platform (Facebook/Meta) shares liability for failing to implement rapid verification and containment during a critical electoral window (7.2).

Procedural Mechanism

6.2 (architectural human confirmation): Social media platforms must implement architectural validation during election windows. The platform must be technically configured to flag and slow the distribution of biometric media of political candidates unless it contains a verifiable, cryptographic authenticity watermark.
7.2 (anti-weaponization): Base model developers are exonerated if they implemented mandatory voice-enlistment verification and unalterable audio watermarking; full liability remains with the hostile actors.
7.1 (independent forensics): The audio stream, metadata signatures, and social propagation footprints must be logged in a secure ledger for national security attribution.