Evidence Pack: The Real Impact of AI Deepfakes on Recent Global Elections

The buildup to the 2024-2025 global election super-cycle was accompanied by a singular, pervasive fear: that generative artificial intelligence would break democracy.[6]

Pundits and policymakers warned of a "deepfake tsunami"—highly realistic, AI-generated audio and video that would brainwash voters, swing tight races, and plunge the electorate into an alternate reality.[3]

Now, with the data tallied from over 60 national elections across the globe, a consensus has emerged among threat intelligence analysts and election security researchers: the worst-case scenarios did not materialize.[6]

According to comprehensive post-election reviews, while AI tools were widely accessible, they largely failed to achieve the mass deception required to alter electoral outcomes.[1][2]

The Alan Turing Institute's analysis of the European elections, for instance, found that a mere 27 pieces of AI-generated political content achieved viral status.[1]

Microsoft Threat Intelligence corroborated this, noting in their 2025 Digital Defense Report that high-production synthetic videos rarely reached broad audiences or caused genuine confusion.[2]

Instead, researchers found that audiences continued to gravitate toward simple, traditional digital forgeries—cheaply edited photos or out-of-context clips—rather than sophisticated deepfakes.[2]

This is not to say that AI was absent from the electoral landscape. A global analysis by Surfshark revealed that 38 countries experienced election-related deepfake incidents between 2021 and 2025.[5]

High-profile examples included the AI-generated robocall mimicking U.S. President Joe Biden in New Hampshire and synthetic audio of a Slovakian party leader discussing election rigging.[3][5]

However, the Brennan Center for Justice notes that these incidents, while alarming, were quickly detected, debunked by journalists, and mitigated by election officials before they could shift voter behavior at scale.[3]

The true impact of generative AI on recent elections proved to be far more subtle, and in many ways, more insidious than mass brainwashing.[6]

The most significant consequence identified by researchers is the amplification of the "liar's dividend"—a phenomenon where the mere existence of deepfake technology allows politicians to dismiss authentic, damaging evidence as AI-generated.[3]

By weaponizing public skepticism, unscrupulous actors can evade accountability, knowing that voters are already primed to doubt the authenticity of digital media.[3]

Furthermore, while AI failed to disrupt macro-level voting patterns, it proved highly effective as a tool for micro-level, targeted harassment.[4]

Reports from the International Institute for Democracy and Electoral Assistance (IDEA) and the Alan Turing Institute highlight a disturbing trend: the disproportionate use of deepfakes to target female political candidates.[1][4]

In countries ranging from Brazil to the United Kingdom, generative AI was deployed to create non-consensual, explicit imagery aimed at demeaning women in politics and damaging their reputations.[4]

This form of political gender-based violence does not aim to persuade voters on policy, but rather to intimidate female candidates and drive them out of the public sphere entirely.[1][4]

Meanwhile, behind the scenes, the most effective cyber threats to campaigns remained distinctly traditional.[2]

Microsoft's analysis revealed that phishing campaigns and Distributed Denial of Service (DDoS) attacks far outpaced AI-driven influence operations in terms of actual disruption to political infrastructure.[2]

Ultimately, the resilience of the 2024-2025 election cycle offers a crucial lesson: democratic institutions, when paired with an alert media and a skeptical public, are remarkably robust against synthetic media.[6]

Moving forward, the focus of election security must pivot from panic over hypothetical mass deception to addressing the very real challenges of the liar's dividend and the targeted harassment of vulnerable candidates.[3][6]