When Artificial Intelligence Enters the Voting Booth: Deepfakes, Disinformation, and a New Threat to European Democracy

Translated by Federico Emanuele

The electoral consultations held at the end of October in Ireland and the Netherlands marked a turning point in the emergence of generative artificial intelligence as an electoral risk factor. In Ireland, a deepfake video circulated just days before the vote falsely showed candidate Catherine Connolly announcing her withdrawal from the race, accurately mimicking the aesthetics and tone of a public broadcaster RTÉ news segment. The video, later removed by Meta and YouTube, revealed the ability of AI to replicate credible media formats and to interfere at the most sensitive stage of the democratic process: the formation of final voter preferences.

In the Netherlands, the most emblematic case involved two members of parliament from the Freedom Party (PVV), linked to a Facebook page that disseminated deepfake images of Labour-Green leader Frans Timmermans in manipulated scenarios, including fake handcuffs and scenes suggesting financial corruption. The systematic use of AI for political delegitimization forced party leader Geert Wilders to issue a public apology. The Dutch case highlights an emerging trend: the integration of AI into the communication strategies of political parties, particularly among actors whose identities rely on provocation, norm-breaking, and so-called shock value as structural elements of their political style.

Beyond manipulated content, another major risk vector lies in AI-driven conversational tools. The Dutch Data Protection Authority warned voters not to use chatbots for voting advice, noting that such systems tend to generate biased or inaccurate responses, misrepresent party positions, and steer users toward specific political options. In a context where a large portion of the electorate makes its final choice within the last 48 hours, the ability of AI to influence behavior through personalized interactions represents a potentially more systemic risk than visual deepfakes.

At the regulatory level, the European Union has recognized political manipulation through AI as a high-priority risk. The AI Act classifies as “high risk” those systems capable of influencing political behavior or altering opinion formation. This entails obligations for transparency, labeling of AI-generated content, and risk management requirements for high-impact models. However, several academic and institutional analyses indicate a significant gap between the law and its practical enforcement. During the Dutch campaign, most manipulated content lacked visible labeling or included only the automatic disclaimers generated by platforms rather than by the political parties responsible for spreading them. In parallel, the European Commission is drafting new guidelines for high-risk AI systems and preparing a proposal to strengthen the protection of electoral campaigns from foreign interference. Yet, at least in their initial draft, these tools will not include legally binding obligations, which could limit their effectiveness.

Another critical issue concerns the legal definitions of manipulated content. Experts have pointed out that the European definitions of “deepfake” and “substantial manipulation” contain ambiguous elements that leave room for interpretation and could weaken enforcement. This is particularly relevant for distinguishing between legitimate editing and deceptive manipulation, a boundary that AI technology increasingly blurs. Without clearer regulatory definitions, political actors may continue to operate in a legal gray area.

On the societal side, surveys show growing skepticism among European voters toward the use of AI in democratic processes, combined with widespread difficulty distinguishing authentic content from artificial creations. Individuals with lower levels of digital literacy are especially vulnerable, exposed both to visual deepfakes and chatbot-driven misinformation. This generates an information gap that reinforces participatory inequalities and enables targeted manipulation.

To strengthen the resilience of European democracies, a multilayered strategy is required. First, labeling of AI-generated content must become mandatory and standardized, supported by robust and verifiable watermarking tools. Second, clear accountability must be established for political parties and platforms, along with stronger audit mechanisms. Finally, it is crucial to invest in advanced digital literacy programs that equip citizens with critical skills to navigate an increasingly opaque information environment. The October elections may serve as a preview of what could become the new normal in future political campaigns.

The ability of institutions to anticipate abuses, correct vulnerabilities, and enforce clear standards will determine whether AI becomes a neutral support tool for democratic life or a powerful instrument of political manipulation.

Mondo Internazionale APS - All Rights Reserved ® 2025