The rapid development of artificial intelligence (AI) has brought unprecedented convenience and transformation to society—but it has also opened doors to new risks. Among these, the misuse of AI-generated deepfakes has increasingly threatened the integrity of democratic processes worldwide, particularly elections.
As a pivotal global election year, 2025 has already seen AI deepfake interventions across multiple key votes—from the Netherlands and Ireland to elections around the world—sparking widespread concern over potential damage to democratic institutions. The European Union, a global pioneer in AI regulation, has sought to address these risks through its landmark AI Act. Yet questions remain over its practical enforceability and effectiveness.
In the October 2025 European elections, AI deepfake content had tangible effects in several countries. In both the Dutch and Irish parliamentary elections, multiple incidents emerged that exposed the technology’s potential to distort voter perception and undermine trust in democracy.
During Ireland’s presidential election, an AI-generated deepfake video circulated online, impersonating the national broadcaster RTÉ and falsely claiming that candidate Catherine Connolly had withdrawn from the race. The fabricated clip spread rapidly, prompting Connolly to publicly clarify the situation and condemn the act as “a disgraceful attempt to mislead voters and undermine democracy.” The incident highlighted not only the deceptive power of deepfakes but also the potential for chaos and reputational damage at critical moments of the electoral process.
The Netherlands’ parliamentary elections were similarly affected. During the campaign, a Facebook account disseminated AI-generated images attacking specific candidates. Research conducted by the University of Amsterdam and Johannes Gutenberg University Mainz confirmed the penetration of AI content into the Dutch election discourse: of roughly 20,000 election-related social media posts, over 400 were AI-generated, with more than 120 traced to accounts linked to the Liberal Party. Claes de Vreese, Professor of AI and Sociology at the University of Amsterdam, described this as “perhaps the first time a European election has faced a fully integrated AI challenge.” The findings underscore that AI is not merely an isolated disruptor—it can be strategically employed as a tool to influence electoral outcomes.
Success
You are now signed up for our newsletter
Success
Check your email to complete sign up
Beyond deepfake videos and images, AI chatbots are also emerging as subtle influencers. When users ask these tools “who should I vote for,” they may receive recommendations that exaggerate polarization and distort the political landscape. Since many voters make their final decisions just days before polling, this subtle guidance could decisively affect election results. On Oct. 21, the Dutch Data Protection Authority (DPA) warned voters against seeking electoral advice from AI chatbots, citing concerns over such hidden forms of manipulation.
The impact of AI deepfakes on elections is multi-layered. They not only generate false information and confuse audiences, but can also stir emotions and deepen societal divides. At a deeper level, when voters struggle to distinguish fact from fiction, trust in the media, political figures, and the democratic system itself erodes. This collapse of trust poses long-term, profound risks to the stability and functioning of democratic societies.
EU’s AI Act: Ambitious regulation meets practical challenges
In response to the growing threat of AI deepfakes to democratic processes, the European Union officially implemented its groundbreaking Artificial Intelligence Act (AI Act) on Aug. 1, 2024, making it the world’s first comprehensive legal framework for AI regulation. The legislation aims to mitigate risks arising from AI—including AI-driven misinformation—while safeguarding fundamental rights, security, and ethical principles. It classifies AI systems according to risk levels and imposes specific transparency obligations on developers and deployers.
At the heart of the AI Act is a tiered “risk-based” approach, dividing AI systems into four categories: unacceptable risk, high risk, limited risk, and minimal risk. Different regulatory requirements apply depending on the risk level. AI systems “intended to influence election or referendum outcomes” are explicitly classified as high-risk, subjecting providers and deployers to strict obligations such as risk assessments, data quality assurance, human oversight, and transparency reporting.
The Act also applies to highly sophisticated general-purpose AI models (GPAI) like OpenAI’s ChatGPT and Google’s Gemini, requiring them to mitigate “systemic risks” to democratic processes. These models must disclose copyrighted material used in training datasets and ensure transparency in their outputs.
Notably, the AI Act has extraterritorial effect: even if an AI developer or provider is based outside the EU, their systems must comply with the law if outputs are used within the EU. This is designed to prevent regulatory gaps and ensure that all AI products and services operating in the EU market meet EU standards. Violations can result in fines of up to €35 million or 7 percent of global annual revenue from the previous fiscal year, providing a strong deterrent against non-compliance.
Challenges in enforcement and limited deterrence of the EU AI Act
Although the EU’s Artificial Intelligence Act is hailed as a milestone in global AI regulation, it faces significant criticism and practical hurdles in implementation and enforcement. These challenges stem not only from the complexity and rapid evolution of AI technology but also from the design of the regulatory framework itself.
1. Enforcement difficulties
Implementing the law across EU member states is a major challenge. Differences in national regulatory frameworks, coupled with the absence of a single unified oversight body, make cross-border enforcement and transparency compliance difficult. Election periods often leave regulators observing from the sidelines, unable to intervene effectively in real time. Meanwhile, the fast pace of AI innovation means that new deepfake techniques and dissemination channels can quickly outpace current legal coverage. Emerging trends—such as encrypted direct messaging, influencers engaging in political promotion, and opaque digital campaign funding—remain difficult to regulate under existing frameworks.
2. Limited deterrent effect of fines
While the law imposes hefty fines, potentially deterring large corporations, smaller players or entities operating in loosely regulated regions may not feel the same pressure. Moreover, fines are largely reactive, applied after violations occur, rather than preventive measures that curb the creation and spread of deepfakes at the source.
3. Delays in full implementation
Although parts of the Act came into effect in August 2024, full implementation is scheduled for Aug. 2, 2026. Given the rapid development of AI technology, this time gap risks rendering portions of the legislation outdated before it is fully operational.
4. Potential conflicts with other laws
Certain provisions of the AI Act, such as transparency obligations, could conflict with the Digital Services Act (DSA). Generative AI blurs the lines of intermediary services defined under the DSA, creating legal ambiguities and uncertainty in enforcement.
5. High compliance costs
The strict regulatory requirements may impose significant costs, particularly for small and medium-sized enterprises, potentially stifling innovation and competitiveness. Concerns have been raised that excessive regulation could even prompt companies like OpenAI, whose CEO Sam Altman has noted potential withdrawals from the European market, to reconsider operating in the region. Compared with the relatively relaxed regulatory environment in the United States, Europe’s “rule-based” approach may place its AI industry at a competitive disadvantage.
6. Ambiguity in risk classification
The Act classifies AI systems “intended to influence election results” as high-risk, but in practice, proving intent or causation for “significant electoral harm” is extremely difficult. The impact of an AI system depends on user behavior, context, and purpose, complicating consistent risk classification. For example, AI applications used for micro-targeted political ads or deepfake generation may not be deemed high-risk if intent cannot be clearly demonstrated.
7. Resource and capacity constraints
Regulatory bodies, even under frameworks like the DSA, often lack sufficient technical expertise and resources to manage rapidly evolving AI platforms. Large tech companies may also resist EU oversight, further complicating enforcement efforts.
The impact of AI deepfakes on European and global elections is now an irreversible reality. While the EU’s AI Act represents a forward-looking attempt to address these threats, its practical enforceability remains limited. Effectively countering the risks posed by AI deepfakes requires a multi-layered, cross-sectoral strategy that is dynamic, comprehensive, and adaptable to evolving technologies and societal contexts.
Interference by AI deepfakes is not only a technological challenge but also a test of human society’s wisdom and resilience. The EU’s experience offers a valuable lesson: legislation is essential, but effective regulation must be ongoing, flexible, and capable of evolving alongside technological and social change. Only through such an approach can democratic foundations be safeguarded while supporting the development of AI.
By Yang Tianzi
