Deepfakes and democratic elections

"Deepfakes and democratic elections" is the framing that AI-generated synthetic media — fabricated audio, video, and images of political candidates and officials — poses a specific, structural threat to democratic elections: capable of spreading viral disinformation before fact-checkers can respond, suppressing votes through impersonation of trusted political figures, and generating a "liar's dividend" that erodes the evidentiary baseline on which political accountability depends. The framing is distinct from the deepfake-abuse framing (oriented around non-consensual intimate imagery and gender-based violence) — its domain is the integrity of democratic deliberation: the conditions for informed voting, the accountability of political actors, and the epistemic environment of elections. It crystallised as an organised civil-society frame in 2018–2019, entered the mainstream policy register through a wave of high-profile incidents in the 2023–2024 global election supercycle, and by 2024 had generated legislation in multiple jurisdictions and an open-standards movement in the technology industry.

The liar's dividend: the foundational framing

The concept central to the framing was articulated by law professors Bobby Chesney and Danielle Citron in their 2019 paper "Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security", published in the California Law Review. The paper introduced the "liar's dividend" — the perverse epistemic benefit that the existence of convincing synthetic media confers on bad-faith actors in two directions simultaneously: they can fabricate damaging content and present it as authentic video evidence, and they can deny the authenticity of real damaging content by claiming it is synthetic. The liar's dividend concept captures the threat's distinctive character: unlike most forms of disinformation, which produce false positives (false stories believed as true), synthetic media technology also produces false negatives (true stories disbelieved as synthetic), expanding the manipulative surface in both directions. Chesney and Citron identified elections as a primary vulnerability — fabricating a candidate's video concession, creating synthetic audio of an official accepting bribes, or generating realistic footage of a manufactured incident days before polling day could each determine outcomes in ways that traditional corrections regimes (fact-checking, platform content policies, legal prohibitions on false-statement-of-fact) were not equipped to address at the speed the technology enables.

WITNESS and the civil-society framing (2018–2019)

WITNESS, the human-rights media organisation, launched the "Prepare, Don't Panic: Deepfakes and Synthetic Media" initiative in 2018 — intervening before the problem had become a mainstream political concern. WITNESS's distinctive contribution was to resist the framing that deepfakes warranted panic-driven suppression. Its "Prepare, Don't Panic" positioning paired awareness of synthetic-media risks with a systematic defence of human rights defenders who rely on video and audio evidence — people who document atrocities, activists who record police violence, journalists who authenticate official statements. Any content-removal response to synthetic-media threats that was over-broad, under-targeted, or platform-automated without human oversight risked suppressing authentic evidence alongside fabrications. WITNESS developed ethical guidelines for civil society, media organisations, and human rights practitioners on how to assess synthetic media risks in high-stakes contexts without generating a chilling effect on authentic documentation.

By November 2023 — with the first major election incidents already documented — WITNESS Executive Director Sam Gregory testified before the US House Oversight Committee on the state of deepfake technology, arguing for disclosure-based rather than blanket-removal-based legislative approaches and for global coordination to avoid a patchwork of over-broad national laws that could export censorship-compatible regulatory templates to authoritarian states. The "Prepare, Don't Panic" framing remained WITNESS's consistent position as the legislative cycle accelerated: the threat was real and warranted response, but the response required specificity — targeting deceptive synthetic content rather than the underlying technology — to avoid capturing truthful political expression, satire, and parody.

The 2023–2024 election supercycle

The deepfakes-and-democracy framing moved from civil-society advance-warning to mainstream legislative concern following a concentrated cluster of incidents in 2023–2024 — the largest global election cycle in recorded history, in which more than half of the world's population was eligible to vote in national elections.

Slovakia (September 2023)

The incident most frequently cited as the first major documented case of deepfake audio deployed in a national election was the fabricated recording circulated on September 28, 2023 — during Slovakia's legal media moratorium — purportedly of Progressive Slovakia leader Michal Šimečka and journalist Monika Tódová of the newspaper Denník N, discussing buying votes from Roma voters and rigging the election count. The recording spread under Facebook's pre-election quiet period that prevents real-time fact-checking in the 48 hours before election day. Slovakia's election was held on September 30, 2023; the deepfake could not be effectively debunked within the moratorium window. Subsequent research identified links to pro-Kremlin disinformation networks in the recording's origin and distribution.

The incident demonstrated two architectural vulnerabilities simultaneously: that deepfake audio of sufficient quality to mislead a non-expert listener was available to relatively unsophisticated actors, and that platform pre-election quiet periods — originally designed to prevent last-minute political advertising — inadvertently created a window in which viral synthetic media was immune to platform correction. The Slovak case entered the academic and policy literature as the reference incident for the moratorium-window class of synthetic-media vulnerability.

New Hampshire Democratic primary robocall (January 2024)

The New Hampshire Democratic primary robocall of January 21, 2024 was the first confirmed use of AI voice synthesis for voter suppression in a US election. A synthetic voice mimicking President Biden's speech patterns — including his characteristic phrase "what a bunch of malarkey" — was used to send robocalls to an estimated 5,000–25,000 registered New Hampshire Democrats on the eve of the January 23, 2024 primary, telling voters to "save your vote for the November election" and "voting this Tuesday only enables the Republicans." The New Hampshire Attorney General opened a criminal investigation. The Federal Communications Commission subsequently issued a $6 million fine — its first involving generative AI technology — against political consultant Steve Kramer, who commissioned the call while working for a rival Democratic presidential campaign. Kramer was indicted on 13 counts of voter suppression and 13 counts of impersonating a candidate across four New Hampshire counties.

The New Hampshire incident established the voter-suppression use case as empirically real — synthetic audio of a candidate instructing voters not to vote, distributed at election-eve speed and scale, was feasible with commercially available voice-synthesis tools and already had a domestic perpetrator, enforcement record, and criminal prosecution trail. Together with the Slovakia incident, it supplied the case evidence that shifted the deepfakes-and-elections framing from prospective concern to documented pattern.

The broader 2024 cycle

The 2024 election year produced documented synthetic-media incidents across multiple countries — fabricated audio and video in Taiwan's January 2024 presidential election, AI-generated content targeting candidates and parties in India's general elections (April–June 2024), manipulative synthetic content in Indonesia's February 2024 election — establishing the problem as a cross-border structural feature of elections in the generative-AI era rather than a single-jurisdiction anomaly.

Legislative and technical responses

The framing's most direct legislative product is mandatory-disclosure regulation for AI-generated election content. The EU AI Act, enacted June 2024, requires through Article 50 that AI systems generating synthetic audio, video, or image content disclose that the output is artificially generated or manipulated — with explicit application to AI-generated content in political advertising and electoral contexts. Multiple US states enacted targeted election-deepfake legislation in 2023–2025, including Minnesota and California, requiring disclosure of AI-generated political advertising and providing removal mechanisms for materially deceptive synthetic election content distributed without disclosure.

The parallel technical response is the Coalition for Content Provenance and Authenticity (C2PA) — an open standards body co-developed by major technology companies that embeds cryptographically-signed provenance metadata into media files at the point of creation, enabling any downstream application implementing the standard to verify whether content has been AI-generated or manipulated. Major news organisations and technology platforms have adopted the standard's "Content Credentials" framework. C2PA represents the technical infrastructure side of the framing's response: where legislation addresses the production and distribution of deceptive synthetic content, content credentials make the authentic-vs.-synthetic distinction verifiable rather than merely perceptible, providing a positive-authentication mechanism that complements removal and penalty regimes.

The civil-society tension: overbroad bans and free expression

The framing's legislative success has generated a counter-tension that its civil-society originators anticipated. South Korea enacted a blanket election-period prohibition on AI-generated audio-visual content used for election campaigning in December 2023 — criminalising the production, editing, distribution, or posting of AI-synthesised content for election-period campaigning irrespective of the content's truthfulness, the consent of any depicted person, or the presence of disclosure labels, during the 90-day window before election day. The Korea National Election Commission removed 388 election-period deepfakes under the law during the April 2024 National Assembly election. The law is the subject of a constitutional complaint filed by Open Net Korea in October 2025, arguing that the blanket technology-based prohibition captures truthful political expression, satire, parody, and anonymous political speech alongside maliciously deceptive synthetic media — the overbreadth argument that WITNESS had consistently advanced in the international-policy debate.

The Open Net Korea complaint is the corpus's clearest instance of the civil-society tension the framing generates: between the well-documented threat synthetic media poses to elections, and the equally documented risk that legislative responses targeting the technology rather than the harm enable censorship-compatible suppression of protected political speech. The constitutional litigation before the Constitutional Court of Korea will be a precedent of first impression on whether an AI-specific blanket election-speech prohibition can survive constitutional review under a developed democracy's free-expression jurisprudence — a question whose answer will shape the framing's legislative trajectory internationally.

The framing in the Global South

The AfricTivistes campaign on AI, elections, and democratic information integrity in West Africa (2023–ongoing) is the corpus's most developed instance of this framing's propagation into organised civil-society infrastructure in the Global South. AfricTivistes' April 2025 joint study with Democracy Reporting International — finding that no major AI chatbot could be considered a reliable source of electoral information for Senegal's November 2024 elections — extends the framing beyond deepfakes specifically to the broader synthetic-media-and-electoral-information register. Its civic-technology investments (the Farafina.tech elections database, the Soobu/Tech4Elections mapping tool, the EUR 175,000 Election Civic Tech Fund) represent the regional operationalisation of the elections-integrity concern into institution-building and capacity infrastructure rather than primarily into legislative advocacy — a posture that maps directly onto the WITNESS "Prepare, Don't Panic" framing's emphasis on preparation and resilience over prohibition and removal.