South Korea - AI Avatar "AI Yoon Seok-yeol"

What Happened

During the 2022 South Korean Presidential Election campaign, the political party of candidate Yoon Seok-yeol created and deployed a highly realistic AI-generated video avatar named "AI Yoon Seok-yeol." The deepfake avatar used Yoon's actual voice and physical likeness, speaking to voters in short, personalized videos on social media, answering questions, and utilizing humorous, meme-focused language designed to appeal to younger voters. While authorized by the candidate and his party, the deployment of the deepfake avatar was highly controversial, raising significant concerns about the normalization of biometric cloning in democratic campaigns, the potential for manipulation, and the blurring of the line between real and synthetic political representation.

How MEG 1 would have acted

While authorized, the biometric cloning of a high-profile political candidate to represent political stances represents a high-risk cognitive manipulation vector. It compromises the cognitive integrity and informational sovereignty of the electorate.

MEG 2 Analysis - Legal Framework

Agent level: N3 (individualized/advanced) - generative AI model used to produce highly realistic, politically active biometric simulations (deepfakes) used systematically in a national campaign.
Cause (6.1):
(c) biometric impersonation (authorized) - the systematic fabrication of a public persona using generative AI to represent political stances; though authorized, it creates severe systemic risks regarding information verification and public trust.
Attachment of liability: The political campaign as the operator is liable for the representations made by the AI avatar; the developers are liable if they failed to implement mandatory digital watermarking and clear, persistent disclosure (7.2).

Procedural Mechanism

6.2 (architectural human confirmation): The deployment of synthetic political candidates must require architectural disclosure. Platforms must technically block the distribution of deepfake political content unless it contains unalterable, machine-readable cryptographic signatures verifying its synthetic nature.
7.2 (anti-weaponization): Software developers are exonerated if they implemented mandatory, visible watermarks and verified consent; full liability remains with the campaign operator.
7.1 (independent forensics): The generation parameters, the prompt inputs, and the cryptographic signatures of the AI avatar must be logged in a secure ledger for public verification.