By Stuart Kerr, Technology Correspondent, LiveAIWire
AI political deepfakes reached genuine electoral significance for the first time in 2024, and a robocall in New Hampshire showed exactly why. An AI-synthesised voice impersonating President Joe Biden urged voters not to vote in the state’s primary election, a direct attempt to suppress democratic participation using technology that costs almost nothing to deploy at scale. The incident was caught and publicised, but it offered a preview of an information environment in which synthetic political content is cheap, convincing, and nearly impossible to distinguish from authentic material without specialist tools that the general public does not have.
The use of AI to generate political content is not inherently problematic; campaigns have always used technology to communicate at scale. What distinguishes the current moment is the combination of near-perfect voice and image synthesis, micro-targeted distribution through social media platforms, and the speed at which false narratives can be seeded and spread before fact-checkers can respond. The window between the creation of a deepfake and its debunking can be long enough to alter electoral outcomes in tightly contested races.
The Scale of AI Political Deepfakes
The 2024 election cycle was widely described by researchers at Stanford, Harvard, and the Oxford Internet Institute as the first in which AI political deepfakes reached electoral significance. Deepfake audio recordings of political figures in Slovakia, Bangladesh, and the United States were used in attempts to influence voters. In Taiwan, researchers documented a coordinated campaign using AI-generated news anchors to spread disinformation in the months before the January 2024 presidential election.
Research published by the Stanford Internet Observatory documented over 300 incidents of AI political deepfakes used with apparent intent to influence public opinion during the 2024 election cycle across more than 40 countries. The vast majority went undetected by platform moderation systems and received no fact-checking coverage because the volume was too large for existing verification infrastructure to handle.
Detection and the Arms Race
The technical response to AI political deepfakes is a classic arms race. Detection tools improve; generation tools improve faster. Watermarking approaches, which embed invisible signals in AI-generated content, have been proposed as a systemic solution, but they depend on the cooperation of content generators, a dependency that is unreliable when adversarial actors are involved.
Platforms including YouTube, Meta, and TikTok have introduced policies requiring disclosure of AI-generated political advertising, but enforcement is inconsistent and policies apply only to paid advertising, not to the organic distribution of synthetic content through user accounts.
Regulatory Responses
Legislative responses to AI political deepfakes are proliferating but fragmented. The United States has seen a wave of state-level legislation requiring disclosure of AI in political advertising, with California, Texas, and Minnesota among the early movers. The EU’s AI Act includes provisions relevant to deepfakes and synthetic political content, building on the Digital Services Act’s requirements for transparency about algorithmic content curation.
The challenge is that effective regulation requires coordination across jurisdictions, since a disinformation campaign generated in one country and distributed through servers in another to influence elections in a third can evade the reach of any single regulatory authority, a governance gap LiveAIWire has traced in our coverage of AI diplomacy and how nations use code as a form of soft power.
The Liar’s Dividend
The psychological research on synthetic media and belief formation suggests that exposure to deepfakes, even when subsequently corrected, can leave lasting traces in how people evaluate the authenticity of genuine content. This liar’s dividend effect, in which the existence of deepfakes allows bad actors to deny genuine recordings, may be as damaging to democratic discourse as the deepfakes themselves, a dynamic LiveAIWire examined directly in our coverage of how political leaders are leaning on AI as the new fake news. Research published in the journal Nature on the cognitive effects of synthetic media exposure has found that people who have encountered deepfakes become less confident in their ability to evaluate the authenticity of any political media, including genuine recordings.
What This Means for You
The media literacy response to AI political deepfakes faces structural limitations. Encouraging voters to verify political content before sharing it places the burden of managing a systemic problem on individual citizens, each of whom is processing hundreds of pieces of political content during an election campaign. Research on correction effects consistently finds that corrections, even accurate and timely ones, reduce belief in misinformation by less than the initial exposure increased it.
Voters in every democracy should now approach political audio and video content with a scepticism that was not previously necessary. Verification tools including reverse image search, cross-referencing with established news organisations, and checking the publication date and source of viral content meaningfully reduce the risk of sharing synthetic disinformation, though the cognitive burden of navigating a synthetic information environment falls unevenly on citizens, requiring digital literacy skills and time that are not equally distributed across the population. This same uneven capacity to verify algorithmically mediated content runs through LiveAIWire’s coverage of AI border surveillance, where the burden of contesting an algorithmic decision falls hardest on those least equipped to do so.
Structural Solutions, Not Individual Vigilance
Structural responses, whether regulatory, technical, or educational, are needed that do not rely primarily on individual vigilance. The integrity of democratic processes in the AI era depends on getting this right, and the current pace of response is not commensurate with the scale of the threat. Democratic institutions that were slow to respond to the first generation of social media disinformation are now facing a more technically sophisticated challenge with less preparation time and less institutional experience to draw on. The cost of continued inadequacy is not merely policy failure, it is the erosion of the epistemic foundation on which democratic legitimacy depends.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity, and the social impact of emerging technology. He publishes daily at LiveAIWire.com.