Argentina - AI Fabricated Images in the Election Campaign
What Happened
During the 2023 Argentinian Presidential Election campaign, both major political camps utilized generative AI image models (such as Midjourney and DALL-E) to produce thousands of highly realistic, fabricated images depicting their rivals. The campaign of candidate Sergio Massa generated images showing his rival, Javier Milei, as a madman or a caricature of a dictator, while Milei's campaign generated images showing Massa as an authoritarian figure or a thief. These AI-generated images were distributed systematically as political advertisements and social media content, blurring the line between caricature, political commentary, and fabricated visual evidence, creating a highly volatile information environment for voters.
How MEG 1 would have acted
The systematic deployment of generative AI to fabricate visual evidence representing political rivals constitutes a high-risk cognitive manipulation campaign, compromising the cognitive integrity and informational sovereignty of the electorate.
MEG 2 Analysis - Legal Framework
Agent level: N3 (individualized/advanced) - generative AI image systems used to produce large-scale, highly realistic, politically active visual simulations deployed systematically in a national campaign.
Cause (6.1):
(c) cognitive manipulation / reputational weaponization - the systematic use of generative AI to fabricate visual representations of public figures to influence public perception and electoral outcomes.
Attachment of liability: The political campaigns as the operators are fully liable for the representations made by their generated content; the image generation platforms share liability if they failed to implement mandatory digital watermarking (7.2).
Procedural Mechanism
6.2 (architectural human confirmation): Generative image platforms and social networks must implement architectural disclosure. Platforms must technically block the generation or distribution of photorealistic representations of public figures during election windows unless they contain unalterable cryptographic watermarks.
7.2 (anti-weaponization): Software developers are exonerated of liability if they implemented mandatory, unalterable metadata watermarks; full liability is placed on the campaign operators.
7.1 (independent forensics): The generation parameters, prompt inputs, and cryptographic metadata of the campaign assets must be logged in a secure ledger for public verification.