In the seven months between August 2024 and March 2025, Korean police accused 963 people of deepfake sex crimes. Over 93% were teenagers or in their twenties. Telegram groups targeting students from more than 500 schools were uncovered. One Telegram bot had 220,000 subscribers and offered two deepfake images for free, with more available for less than a dollar.
South Korea responded with some of the strictest deepfake laws in the world. Three separate legal frameworks now overlap: criminal law targeting sexual content, election law banning all campaign deepfakes, and the AI Basic Act imposing transparency obligations on AI operators. If you operate AI platforms that serve Korean users or generate synthetic media, all three apply.
What Laws Cover Deepfakes in South Korea?
Three distinct legal frameworks address deepfakes, each enforced by different authorities with different penalties.
1. Act on Special Cases Concerning the Punishment of Sexual Crimes (Article 14-2). Criminalizes creation, distribution, and possession of non-consensual deepfake sexual content. First enacted in 2020; significantly amended in October 2024 and June 2025. Enforced by the National Police Agency.
2. Public Official Election Act (Article 82-8). Bans all deepfake content for electioneering purposes during the 90 days before an election. Enacted December 2023. Enforced by the National Election Commission.
3. AI Basic Act (Article 31). Requires labeling of AI-generated content that is “difficult to distinguish from reality.” In force January 22, 2026. Enforced by MSIT.
Each layer addresses a different harm — sexual exploitation, electoral manipulation, and consumer deception — but all three can apply simultaneously to the same piece of content.
Criminal Law: Deepfake Sexual Content
The Act on Special Cases Concerning the Punishment of Sexual Crimes is the backbone of Korea’s deepfake enforcement. Article 14-2 now carries some of the harshest penalties for synthetic sexual content anywhere in the world.
Current penalties (post-October 2024 amendment):
| Conduct | Maximum penalty |
|---|---|
| Creation, editing, or distribution of non-consensual deepfake sexual content | 7 years imprisonment (raised from 5) |
| Possession, viewing, or storage of such content | 3 years imprisonment or KRW 30 million fine (~$21,000) |
| Using deepfakes to blackmail or threaten minors | Minimum 3 years imprisonment |
| Coercing minors into acts using deepfake threats | Minimum 5 years imprisonment |
The October 2024 amendment made two critical changes. First, it criminalized mere possession — previously, only production and distribution were offenses. Second, it removed the requirement to prove intent to distribute before prosecution. Possession alone now triggers criminal liability.
A further June 2025 amendment expanded undercover investigation powers. Officers can now conduct active investigations using false identities in deepfake cases involving adult victims, with legal indemnity for unavoidable investigative actions.
Enforcement in practice: In the eleven months between November 2024 and October 2025, police apprehended 3,557 individuals for cyber-sexual violence. Deepfake-related crimes were the largest single category: 1,553 cases. Nearly 62% of suspects were teenagers. Police submitted over 36,000 removal requests and referred 28,000 victims to the digital sex crime support center.
The landmark conviction came in October 2024: two men who created nearly 2,000 degrading AI-generated images of Seoul National University alumni received sentences of up to 10 years imprisonment — the heaviest penalty for deepfake crimes since their criminalization.
The Crisis That Drove the Law
Korea’s deepfake legislation did not emerge from abstract policy deliberation. It was driven by two waves of public crisis.
The first was the Nth Room case (2020). Perpetrators used Telegram to lure women and girls with false employment offers, then coerced victims into sending intimate images that were weaponized for blackmail and sale. Over 100 deepfake videos and 5,000 images were created. Victims included minors who were forced to label themselves “slaves.” The case triggered the original 2020 criminalization.
The second wave hit in August-September 2024 when investigators discovered Telegram groups targeting students from over 500 middle and high schools. AI-generated sexualized images of students — mostly girls — circulated in encrypted “Humiliation Rooms.” Perpetrators were typically classmates. The abuse was gamified: users earned rewards for inviting friends and escalating harm. This wave drove the October 2024 amendments criminalizing possession.
A 2024 government study found 35.8% of Korean women had suffered a violent act at least once. In 2024, 80% of those apprehended for digital sex crimes were minors or teenagers — and 25% were under 14, placing them below the age of criminal responsibility.
Election Deepfakes: A Complete Ban
The Public Official Election Act, Article 82-8 (enacted December 2023), takes a different approach. It does not distinguish between harmful and benign deepfakes. All deepfake content for electioneering purposes is banned during the 90-day pre-election period, regardless of whether the content is true, false, labeled, or consented to.
Penalties reach up to 7 years imprisonment or fines of KRW 10-50 million. Both creators and distributors face prosecution.
During the June 2025 presidential election, the National Election Commission deployed “Aegis,” a deepfake detection AI model developed by the National Forensic Service and Korea Electronics Technology Institute. Police investigated 6 deepfake videos and 14 suspects targeting Democratic Party candidate Lee Jae-myung.
The blanket ban faces a legal challenge. In October 2025, Open Net Korea filed a constitutional complaint arguing Article 82-8 unconstitutionally restricts political speech — including truthful content, satire, and anonymous expression — based solely on the technology used, not the harm caused. The case is pending before the Constitutional Court.
AI Basic Act: The Transparency Layer
The AI Basic Act (Article 31, in force January 22, 2026) adds a third obligation specifically targeting synthetic media that could deceive.
Article 31(3) requires that when an AI system produces outputs “difficult to distinguish from reality,” the operator must label such content clearly, in a manner users can “readily recognize.” For video deepfakes, the AI-generated nature must be indicated throughout playback — a continuously displayed logo, for example. If content is exported, downloaded, or shared, labeling must be applied to the output itself; metadata-only approaches are insufficient.
An exception exists for artistic and creative works, where labeling may be presented non-intrusively.
Within one month of the Act taking effect, Naver began applying “byCLOVA” watermarks across its services. Kakao introduced automatic watermarking for AI-generated images shared via KakaoTalk. Google applied watermarking by default in its AI video generation services.
Administrative penalties under the AI Basic Act are lighter than the criminal framework: up to KRW 30 million (~$21,000) for failing to label AI outputs or appoint a domestic representative. A one-year grace period applies for enforcement.
How All Three Laws Interact
For AI platform operators, the three laws create overlapping obligations:
| Scenario | Criminal Law applies? | Election Law applies? | AI Basic Act applies? |
|---|---|---|---|
| AI generates sexualized image without consent | Yes — up to 7 years | No (unless election context) | Yes — must label |
| AI generates political deepfake during campaign | Possibly (if sexual) | Yes — banned outright | Yes — must label |
| AI generates realistic but non-sexual, non-political content | No | No | Yes — must label |
| User possesses AI-generated sexual content | Yes — up to 3 years | No | No (user, not operator) |
The practical challenge is that these laws are enforced by three different bodies (police, NEC, MSIT) with different standards, different penalties, and different enforcement postures. Platform operators need compliance programs that address all three simultaneously.
How Does Korea Compare to China and the EU?
| Dimension | South Korea | China | EU |
|---|---|---|---|
| Deepfake sexual content | Criminal — up to 7 years | Prohibited under general “illegal content” rules | Not specifically addressed (civil/regulatory) |
| Labeling trigger | AI outputs “difficult to distinguish from reality” | All AI-generated content; explicit labels for high-confusion categories | Deepfakes specifically (deployers); all AI outputs (providers, machine-readable) |
| Who bears labeling obligation | AI business operators | Providers + platforms + app stores | Providers (technical) + deployers (visible) |
| Criminal sanctions | Yes — up to 7 years for sexual content; 7 years for election deepfakes | Referral to criminal law possible | No criminal sanctions (administrative fines only) |
| Platform takedown powers | KCSC can order deletion/blocking | CAC can order removal, suspend services | Not established at EU level (member state frameworks) |
| In force | 2020 (criminal); Oct 2024 (amended); Jan 2026 (AI Basic Act) | Jan 2023 (deep synthesis); Sep 2025 (labeling measures) | Aug 2026 (AI Act Art. 50) |
China’s deep synthesis rules (in force January 2023) require irremovable technical marks on all AI-generated content — more technically prescriptive than Korea’s approach. China also requires real-identity verification before users can publish synthetic content. The CAC actively enforces these rules.
The EU AI Act (Article 50, effective August 2026) requires deployers to disclose deepfake origins and providers to mark outputs in machine-readable format. Unlike Korea, the EU relies exclusively on administrative penalties — no criminal sanctions for deepfake creation. Unlike China, the EU offers a lighter-touch exception for artistic and satirical content.
Korea stands alone in treating deepfake sexual content as a serious criminal offense with prison sentences comparable to physical sex crimes. No other major jurisdiction comes close on penalties.
What Should You Do This Week?
If you operate an AI platform serving Korean users:
- Implement content labeling. The AI Basic Act requires all realistic AI-generated content to carry labels throughout playback. This applies to any platform accessible in Korea, including foreign operators above the domestic representative threshold.
- Build a content moderation pipeline for sexual deepfakes. Korean police are actively investigating and have agreements with Telegram for data sharing. Platforms face both criminal referral risk and KCSC takedown orders.
- If your platform enables political content generation: Understand the 90-day election blackout period. The next presidential term ends in 2030, but by-elections and local elections occur regularly. The blanket ban applies regardless of labeling or consent.
- Monitor the constitutional challenge. If the Constitutional Court strikes down Article 82-8’s blanket election ban, the legal framework for political deepfakes will change significantly.
- Review your terms of service. Korean enforcement increasingly targets both creators and platforms. Clear ToS language on prohibited synthetic content, combined with technical controls, strengthens your legal position.
For background on Korea’s broader AI regulatory framework, see our AI Basic Act guide and PIPC enforcement analysis. For China’s parallel deepfake rules, see our deep synthesis article.
Disclaimer: This content is for informational purposes only and does not constitute legal advice.
Last verified: April 10, 2026
Compare: EU vs South Korea
For the global keystone comparison across twelve dimensions — high-impact vs high-risk classification, mandatory vs voluntary conformity, KRW 30M vs €35M penalties, Korea’s innovation chapter, and a five-step dual-market compliance baseline — see EU vs South Korea AI Act: High-Impact vs High-Risk Compared (2026).