Introduction

During Moldova’s parliamentary election in September 2025, Russian influence operations deployed AI-generated images and voiceovers alongside fake social-media accounts, fabricated news websites and coordinated amplification networks (G7 Rapid Response Mechanism, 2026). The election took place in the midst of deep divisions between pro-European and pro-Russian political parties, with its outcome carrying significant implications for Moldova’s future regarding the integration to the European Union (OSCE Office for Democratic Institutions and Human Rights, 2026). By April 2026, post-election assessments provided a clearer picture of the role generative AI played in these foreign operations.

Moldova is an instructive case because generative AI was used alongside more established methods of information operations (G7 Rapid Response Mechanism, 2026; Promo-LEX, 2025). Evidence suggests that AI expands the capacity to produce manipulative content, while conventional distribution methods remain important for achieving wider reach and amplification (Bentzen, 2025; G7 Rapid Response Mechanism, 2026; Williams et al., 2025).

Documented cases of AI-enabled election interference 

Joint election monitoring by the G7 Rapid Response Mechanism (RRM) detected more than 450 pieces of foreign information manipulation and interference (FIMI) content targeting Moldova’s election. During its first monitoring phase, Russian state-aligned operations accounted for 91 percent of the identified FIMI content, with activity peaking during the election weekend (G7 Rapid Response Mechanism, 2026).

Operation Undercut, a Russian state-aligned influence campaign linked to the Social Design Agency, Structura and ANO Dialog, illustrates how generative AI was used to produce political content. The operation distributed videos on X and TikTok using AI-generated voiceovers and images alongside screenshots of mainstream media, Telegram posts and infographics (G7 Rapid Response Mechanism, 2026).

Separately, Moldovan election monitor Promo-LEX identified another coordinated TikTok network of 25 core accounts and almost 500 supporting accounts sharing identical videos, graphics and narratives, including AI-generated content. They assessed the network as an effort to artificially amplify political narratives and create an impression of widespread public support. This example illustrates how synthetic content can be widely distributed through coordinated social-media activity, which, in itself, is not a new method.

Storm-1516, a Russian influence operation tracked under this name by Microsoft's Threat Analysis Center (MTAC), illustrates another established distribution method, namely the use of fabricated local media outlets. The operation established at least 45 websites posing as Moldovan or Romanian news organizations in the weeks preceding the election. It used inauthentic websites to share fabricated stories and published videos assessed as likely AI-generated before amplifying the content through social-media accounts (G7 Rapid Response Mechanism, 2026).

These cases were all different in both their methods. Operation Undercut focused on AI-generated audiovisual content, the Promo-LEX identified network used hundreds of coordinated TikTok accounts to amplify narratives and Storm-1516 relied on fabricated local news websites to distribute misleading stories. All together, they illustrate how AI-generated content was incorporated into largely already established methods of information manipulation.

Easier production doesn’t equal wider reach

Researchers have found that contemporary LLMs can generate high-quality disinformation content at substantially lower cost than traditional methods, including in hyperlocalised scenarios (Williams et al., 2025). A 2025 European Parliamentary Research Service assessment similarly concluded that generative AI can make content creation and some information-manipulation activities easier and cheaper to conduct (Bentzen, 2025).

However, greater content production does not necessarily translate into greater audience reach. The G7 RRM found that Operation Undercut generated substantial quantities of content but attracted limited engagement. Storm-1516, meanwhile, produced less material, although some of its content exceeded one million views (G7 Rapid Response Mechanism, 2026). These figures are not directly comparable measures of effectiveness, as the operations differed in their distribution methods. This makes it difficult to isolate the contribution of any single factor to their observed reach. Nevertheless, the contrast clearly illustrates that producing more content does not automatically result in greater visibility.

Neither should online reach be treated as evidence of electoral persuasion, because views, shares and comments indicate different forms of exposure or engagement but do not establish whether content changed political attitudes or voting behaviour. The Centre for Emerging Technology and Security’s (CETaS) analysis of the 2024 UK, French and European Parliament elections found no evidence that the AI-enabled disinformation cases it examined affected election results; most exposure instead reinforced existing political beliefs (Stockwell, 2024).

These findings suggest that generative AI can expand the capacity to produce manipulative political content, while established distribution networks remain important for achieving a wider audience (G7 Rapid Response Mechanism, 2026; Williams et al., 2025). The Moldovan case therefore illustrates the need for examining how synthetic content is distributed, rather than focusing solely on its production.

Countering interference through network monitoring

The case of Moldova suggests that responses to foreign interference in elections must address both synthetic content and the networks distributing it. While identifying AI-generated content can help expose individual cases of manipulation, it does not necessarily reveal the wider network through which that content is distributed. Election-monitoring bodies should therefore establish dedicated FIMI-monitoring teams during election periods, tasked with mapping coordinated accounts, distribution structures and repeated narratives across platforms and publicly reporting identified networks.

Promo-LEX’s analysis of coordinated TikTok accounts and the G7 RRM’s monitoring of activity across different social media platforms similarly demonstrate the importance of analysing amplification networks alongside individual pieces of content (G7 Rapid Response Mechanism, 2026; Promo-LEX, 2025).

Conclusion

Moldova’s 2025 election illustrates how generative AI can expand the production of manipulative political content without replacing already established methods of distribution (Bentzen, 2025; G7 Rapid Response Mechanism, 2026; Williams et al., 2025). The case highlights the importance of dealing not only with synthetic content itself, but also with the wider FIMI infrastructure through which it reaches audiences (G7 Rapid Response Mechanism, 2026; Promo-LEX, 2025).