AI-Generated Media Reportedly Used in Russian Disinformation Campaign in Moldova

Russian-linked entities allegedly deployed AI-generated images and videos to spread disinformation aimed at swaying Moldova’s referendum on E.U. membership. The AI-enhanced media campaign included fabricated stories and doctored visuals, the purpose of which was reportedly designed to amplify fear and undermine pro-E.U. sentiment in the days leading up to the referendum.

Risk classification

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

The incident involves a coordinated, large-scale disinformation campaign using AI-generated content to manipulate public opinion and influence a national democratic referendum.

Additional risk subdomains

• 4.3 Fraud, scams, and targeted manipulation: The campaign featured a targeted deepfake video of Minister Dumitru Alaiba designed to humiliate and discredit him personally.
• 3.2 Pollution of information ecosystem and loss of consensus reality: The campaign flooded social media with artificially generated images, videos, and fake documents, polluting the information ecosystem during a critical election.

Causal factors

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

The incident was caused by the intentional decisions and actions of human malicious actors deploying generative AI tools post-deployment to influence an election.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The incident involves the creation and dissemination of AI-generated content such as deepfakes and synthetic media, which requires transparency obligations under the Act.

AI system and alleged parties

• AI system: unspecified
• AI purpose: Deepfake Video Generation; Social Media Content Generation
• Behaviour type: Tool
• Alleged developer: Unknown AI developers
• Alleged deployer: Russia-backed influencers, Maria Zakharova, Ilan Shor, Government of Russia
• Alleged harmed parties: Pro-EU Moldovans, Moldovan general public, Maia Sandu, Government of Moldova, Electoral integrity, Dumitru Alaiba, Democracy

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Substantial. 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 Substantial, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Minor, indirect Negligible
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Substantial, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

Reported: The report explicitly describes the creation and spread of malicious and toxic content, including a deepfake video targeting a minister and fake government letters.

Directly caused: Russian-linked actors directly created and spread a deepfake video of Minister Dumitru Alaiba and fabricated Ministry of Culture letters on social media.

Inferred additional harm: It is likely that further manipulative and toxic synthetic content was generated and circulated across Moldovan social media channels during the campaign.

Democracy

Reported: The report explicitly describes a coordinated campaign aimed at interfering with Moldova's democratic referendum on EU membership.

Directly caused: The campaign directly targeted the Moldovan electorate with fake content to manipulate public opinion and influence the referendum outcome.

Inferred additional harm: The campaign likely eroded trust in democratic institutions and the integrity of the electoral process among the Moldovan public.

Epistemic

Reported: The report explicitly describes epistemic harm through a torrent of disinformation, including bogus news stories, artificially generated images, and deepfake videos.

Directly caused: AI-generated deepfakes and fabricated documents were spread to mislead voters about EU integration and government officials.

Inferred additional harm: The widespread dissemination of fakes likely degraded the shared information environment and made it harder for citizens to distinguish real news from fabricated content.

People affected

• Occurrences reported: 1
• People reportedly harmed: 1
• People reportedly exposed: 1000000

Potential causes

Management

• Neglect of Smaller Regions: Tech firms underinvested in safety and moderation for Moldova.

Technology

• Artificially Generated Media: Deepfakes and AI videos were used to discredit political targets.
• Automated Bot Amplification: Bots and fake accounts rapidly spread synthetic content online.

Data Inputs

• Unverified Synthetic Media: Synthetic videos and images were fed directly into social feeds.
• Inauthentic Account Data: Fictitious profiles provided misleading inputs to platform algorithms.

Human Factors

• Susceptibility to Fakes: Targeted demographics easily believed divisive, fabricated stories.
• Limited Digital Literacy: Users struggled to distinguish AI-generated fakes from real media.

Process and Methods

• Slow Content Moderation: Platforms failed to swiftly block coordinated inauthentic networks.
• Inadequate Detection Tools: Algorithmic moderation struggled to identify localized deepfakes.

Regulatory Environment

• Weak Platform Accountability: Lack of strict regulations allowed foreign disinformation to thrive.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide clear details about the nature of the disinformation campaign, the specific deepfakes used (such as the minister's video), and the involvement of major tech companies like Microsoft and Meta in tracking and mitigating the threat. The role of AI as a tool for generating synthetic media is explicitly stated.
• Ambiguities identified: The specific AI models or platforms used by the malicious actors to generate the deepfakes and images are not named, though Microsoft tracked the campaign.

Russian-linked entities deployed AI-generated deepfakes and fabricated documents in Moldova to influence an upcoming EU membership referendum. While the campaign represents an evolved capability in foreign information warfare and targeted sovereign democratic processes, its overall national security impact remains substantial but manageable through standard counter-disinformation procedures.

• Overall national security impact: Substantial
• Scope: Single nation
• Primary target: Moldova
• Alleged perpetrator: Russian-linked entities and Ilan Shor

Threat characteristics

• Imminence: Near-term. The campaign was active and developing in the lead-up to the October 2024 referendum, requiring ongoing monitoring and mitigation.
• Autonomy: Human-controlled. AI tools were strictly used by human operators to generate media, with no autonomous decision-making or execution by the AI itself.
• Novelty: Evolved capability. While foreign election interference is an established threat, the integration of highly realistic deepfakes of government ministers represents an evolved capability.

Impact by dimension

• Physical security: Negligible. No physical threat, kinetic attacks, or infrastructure damage occurred or were threatened during this incident.
• Information security: Substantial. Russian-linked actors executed a coordinated information operation using AI-generated deepfakes and synthetic media to influence public opinion and target government officials.
• Sovereignty: Substantial. The campaign directly targeted a critical national referendum on EU membership and fabricated official government correspondence, threatening sovereign democratic processes.
• Economic security: Negligible. No critical economic systems, trading algorithms, or strategic technologies were targeted or compromised.
• Societal stability: Minor. The operation attempted population-scale social manipulation to stoke fear, though standard platform moderation and government denouncements limited the overall impact.