Open Net Korea constitutional complaint against the Public Official Election Act deepfake prohibition (2025–ongoing)

On 1 October 2025, Open Net Korea — the Seoul-headquartered civil-society organisation that is the principal Korean anchor on freedom of expression online, net neutrality, intermediary liability, and emerging AI policy — filed a constitutional complaint at the Constitutional Court of Korea challenging Article 82-8(1) of the Public Official Election Act, the December 2023 statutory provision that since 29 January 2024 has prohibited the production, editing, distribution, screening, or posting of "virtual sounds, images, or videos that are difficult to distinguish from reality and created using artificial intelligence technology" for election-campaigning purposes during the 90-day window from ninety days before election day through election day. The complaint is the corpus's first mapped civil-society constitutional challenge to a generative-AI-specific election-speech prohibition, and the principal civil-society response to the wave of "blanket deepfake ban" legislation that has spread across East and Southeast Asia in 2023–2025 in the name of electoral integrity. Open Net's constitutional theory is that Article 82-8(1) regulates by reference to the technology used to create an expression rather than by reference to any harm caused by the expression, that it captures truthful political expression, satire, parody, anonymous political speech, and policy promotion alongside the maliciously deceptive synthetic media the law's drafters claimed to target, and that the prohibition therefore impermissibly burdens the freedom of political expression that the Korean Constitution treats as a component of the free democratic basic order.

The statutory regime

Article 82-8 was inserted into the Public Official Election Act by an amendment passed by the Korean National Assembly on 28 December 2023 and entered into force on 29 January 2024, one month after enactment under the amendment's supplementary provisions. The provision is titled "Election Campaigning Using Deepfake Videos and Similar Methods", and its substantive paragraph 1 prohibits, for the period from 90 days before election day through election day itself, the "production, editing, distribution, screening or posting of virtual sounds, images or videos that are difficult to distinguish from reality, using artificial intelligence technology or other means for election campaigning". The corresponding criminal-penalty schedule under Article 255 sets violations at up to seven years' imprisonment or fines of 10–50 million won (roughly USD 7,500–37,500 or EUR 6,800–34,000). The amendment carves out a narrow exception for AI-generated content explicitly labelled as such and not intended for election campaigning, but otherwise applies to the full range of AI-generated audio-visual material irrespective of its truthfulness, the consent of any depicted person, or the presence of disclosure labels — the four content-neutral axes on which Open Net's constitutional argument is built.

The legislative history identifies the precipitating incident as a late-2022 deepfake video falsely showing President Yoon Suk-yeol endorsing a county mayoral candidate during a municipal-election cycle, which circulated widely on Korean social media and drew calls in the National Assembly for AI-specific election-speech legislation distinct from the Public Official Election Act's existing prohibitions on factual misrepresentation of candidates. The December 2023 amendment was the legislative output of that pressure, and was structured as an AI-specific prohibition layered on top of the Act's pre-existing false-statement-of-fact provisions rather than as an extension of those provisions.

The 2024 election: the law in operation

The 22nd National Assembly election of 10 April 2024 was the first general election conducted under Article 82-8(1) and is the principal record of how the law operated in practice. The National Election Commission established a dedicated AI-identification team on 11 January 2024 — the first day of the 90-day prohibition window — and added two to three AI monitoring agents at each city and provincial election commission. By 16 February 2024 the NEC reported that it had identified 129 deepfakes in violation of the Act since the law's 29 January effective date and that most had subsequently been removed from social media and online platforms; The Readable's later cumulative tally across the full election cycle reported 388 election-period deepfakes flagged or removed. The NEC's anti-deepfake task force comprised 72 officials and experts at the height of the election period.

The 2024 enforcement record is the empirical predicate for Open Net's overbreadth argument. The published NEC takedown counts do not distinguish between deceptive deepfakes (impersonations of candidates uttering statements they did not make), labelled satirical or parodic synthetic media, and AI-assisted production of otherwise truthful political content; nor do the published takedown figures distinguish between content for which the depicted person consented and content for which consent was absent. Open Net's complaint argues that this content-blind operational pattern — under which the NEC has discretion to flag and remove any AI-generated election-period speech irrespective of its harm profile — is the predictable consequence of a statutory text that regulates by technology rather than by harm, and that the Diplomat's identification of the 2024 Korean election as the first major test of an AI-specific election-speech prohibition under live election conditions accordingly carries with it the first major test case of the free-expression cost of such a regime.

Open Net's constitutional arguments

Open Net's complaint advances four substantive constitutional grounds on which Article 82-8(1) is said to be unconstitutional.

First, overbreadth: the prohibition captures the production, editing, distribution, screening, or posting of AI-generated election-period speech irrespective of the content's truthfulness, irrespective of the consent of any depicted person, and irrespective of whether the synthetic origin of the material is clearly labelled — and is therefore "an excessive and unconstitutional regulation unprecedented anywhere in the world" in the complaint's characterisation. Open Net argues that the regulation impermissibly captures four categories of constitutionally protected political speech: truthful political expression and factual reconstruction, satire and parody criticising candidates, anonymous political expression protecting minority voices, and expressions promoting candidate policies.

Second, technology-based discrimination: Article 82-8(1) regulates speech "simply because they were vividly expressed using artificial intelligence technology", treating the medium of expression as itself constitutive of the regulated harm. Open Net's argument — anchored in the position that "technology used in expressive acts is merely a tool or means" — is that the Korean Constitution does not permit a content-neutral targeting of a technology of expression as a substitute for content-based targeting of demonstrable harm. The same proposition runs through Executive Director Kyung-sin Park's parallel Global Free Speech Summit Nashville presentation in October 2025 and supplies the doctrinal backbone of Open Net's broader public-law critique of the technology-targeting register of recent Korean AI legislation.

Third, violation of the principle against excessive restriction: under the Korean Constitution's Article 37(2) proportionality framework, even a fundamental right may be restricted by law where necessary for national security, the maintenance of law and order, or public welfare, but the restriction may not impair the essential aspect of the right. Open Net argues that Article 82-8(1)'s blanket prohibition of an entire mode of expression during a 90-day window fails the proportionality test on multiple axes — the existing Korean statutory framework already addresses demonstrably false election-speech through long-standing false-statement-of-fact provisions, defamation law, and the Information and Communications Network Act, and the AI-specific prohibition therefore captures protected expression without contributing additional regulatory value beyond what the harm-based regime already supplies.

Fourth, the free-democratic-basic-order argument: Open Net positions freedom of political expression as a component of the free democratic basic order that the Korean Constitution requires the state to preserve, and characterises the AI-specific election-period prohibition as an erosion of the democratic-deliberation conditions the Constitution treats as foundational. The argument supplies the constitutional-theory frame within which the more technical overbreadth and proportionality arguments operate.

Place in Open Net's wider AI-and-free-expression record

The constitutional complaint is the litigation-front expression of an Open Net public-law programme on AI and free expression that has consolidated rapidly during 2024 and 2025. The parallel critique of the Korean Basic Act on the Development of Artificial Intelligence — passed by the National Assembly in December 2024 and entered into force on 22 January 2026 — runs the same technology-targeting-versus-harm-targeting argument at the statutory rather than constitutional level: Park's Nashville presentation sets out Open Net's principal critique that the AI Basic Act "makes the government's role prominent in vetting high impact AI, which opens up a possibility for government censorship" and that "these existing laws burden this first amendment activity just because it is carried out with a particular type of software". The eight-point critique advanced at Nashville — most relevantly, the framing of Korea's deepfake regulations as "knee-jerk reactions to unfounded fear" and the position that AI laws "burden First Amendment activities by targeting probabilistic software rather than regulating AI itself" — supplies the international-comparative scaffolding for the constitutional complaint, and the March 2024 APC interview records the same arc as a strategic priority of the organisation's second decade.

The complaint also extends a Korean constitutional-litigation pattern that Open Net has practised since its 2013 founding: from the 2013 constitutional complaint against identity-verification mandates for minors through the cumulative challenges to the internet real-name regime that produced the post-2012 line of Constitutional Court wins, the organisation has used the constitutional-complaint procedure as its central instrument against Korean executive and legislative content regulation. The 2025 deepfake-prohibition complaint is the AI-policy continuation of that arc.

Significance for the broader AI-good movement

The campaign is the corpus's clearest civil-society constitutional challenge to a generative-AI-specific election-speech prohibition, and the principal civil-society response in the corpus to the regional wave of "blanket deepfake ban" legislation that has spread across East and Southeast Asia in 2023–2025. It is the East-Asian counterpart, at the constitutional-litigation register, to the European welfare-algorithm constitutional-litigation campaigns the corpus already tracks — the Panoptykon Foundation challenge to algorithmic profiling of the unemployed in Poland closed by the Constitutional Tribunal's K 53/16 judgment of 6 June 2018, and the parallel French and Dutch welfare-algorithm-litigation registers — and is the most-developed mapped instance of a civil-society organisation positioning the constitutional-litigation register as the primary response to AI-specific speech regulation rather than to AI-specific public-administration deployment. The Constitutional Court of Korea's eventual disposition of the complaint will be a precedent of first impression on whether, and on what constitutional grounds, an AI-specific blanket prohibition of election-period speech is permissible under the Korean Constitution; for the field as a whole, the case is one of the principal in-flight tests of whether the technology-targeting register of recent AI legislation can withstand constitutional review under a developed democracy's free-expression jurisprudence.