AI Regulation Tracker  /  Regulation in force

South Korea bans undisclosed AI election deepfakes in the 90-day window and mandates labeling outside it

Article 82-8 of Korea's Public Official Election Act prohibits realistic AI-generated campaign audio, images and video within 90 days of an election, and requires a clear "AI-generated" label at all other times. Any firm producing Korean political content now carries a live disclosure duty backed by criminal referrals.

By Anthony Guerriero  ·  Reviewed by The Leveraged Years Editorial Desk  ·  Published July 7, 2026  ·  Last updated July 7, 2026

South Korea now runs one of the world's more direct legal controls on AI in elections. Article 82-8 of the Public Official Election Act, added by a December 2023 amendment and known in Korean as the 딥페이크영상등 provision, prohibits anyone from making, editing, distributing or displaying realistic AI-generated audio, image or video for election campaigning within 90 days of election day. The prohibition is not a disclosure rule during that window. It is a ban on the synthetic media itself.

Outside the 90-day window, the law shifts from prohibition to transparency. AI-generated campaign content is permitted, but it must carry a clear label identifying it as AI-generated virtual content. The exact form of that label is set by National Election Commission rule (중앙선거관리위원회규칙) rather than left to the producer, which means compliance is measured against a fixed standard, not a good-faith effort. The design choice matters for anyone building an approval process. Because the standard is fixed by the Commission, a producer cannot argue that a home-grown watermark or an ambiguous caption was reasonable. The label either matches the prescribed form or it does not.

How enforcement actually works

The National Election Commission (중앙선거관리위원회) enforces Article 82-8, working with the Korea Communications Commission and the police on detection and takedown. The scale of activity in the most recent presidential cycle shows the rule is operational, not symbolic. Deepfake-related takedown requests rose from 388 during the 2024 general election to 10,510 during the 2025 presidential election, according to Commission figures reported in the Korean press. That is a jump of more than 25 times in a single cycle.

The Commission also moved from takedowns to criminal referrals. It filed its first criminal complaints against roughly three operators. One was a YouTuber who repeatedly posted an image depicting a candidate in a prison uniform and circulated AI-anchor fake-news videos. The move from removal requests to named criminal complaints marks the point at which the provision stopped being a content-moderation lever and became an enforced criminal-adjacent duty. The prison-uniform image is a useful example because it shows what the ban targets: a realistic fabricated depiction of a real candidate, pushed repeatedly, during the period the statute treats as most sensitive. A takedown clears the content, but a complaint puts the operator on the record.

What the law does not do

The reach of Article 82-8 has clear edges worth stating plainly. It is election law, so it governs content made for campaigning, not synthetic media generally. It does not replace Korea's separate horizontal AI transparency duties, which sit in the AI Basic Act and apply across sectors. And the 90-day ban is time-bound: the same realistic synthetic clip that is prohibited inside the window may be lawful outside it, provided the AI-content label is attached. Producers who treat the labeling regime as a year-round safe harbor will misread the statute, because inside the 90-day window there is no label that cures a realistic deepfake. The correct mental model is two regimes stacked on a calendar: prohibition near the vote, mandatory labeling the rest of the time.

The cross-border angle for US readers

For a US professional, the contrast is the story. The United States has no single federal answer here. The Federal Election Commission has largely declined to act, and roughly twenty states have passed their own deepfake-in-elections statutes, producing a patchwork of definitions, windows and labeling formats. Korea offers the opposite structure: one national ban paired with a single national labeling mandate, and a regulator willing to file criminal complaints. US political-media, advertising and platform-compliance counsel advising on cross-border campaigns, or on Korean-language diaspora content that could touch a Korean election, cannot assume a US-style disclosure posture satisfies Article 82-8. The Korean rule is stricter at its core, because near an election it forbids the content outright rather than asking for a disclaimer.

The practical consequence is a workflow change, not just a legal note. Any team shipping Korean election content needs to know two dates for every election: the polling date and the 90-day line before it. Inside that line, realistic AI-generated campaign media is off the table. Outside it, the label goes on. The 2025 presidential election, with its 10,510 takedown requests and first criminal complaints, is the evidence that the Commission treats both halves of the rule as live.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.