Correction (19 October 2026): An earlier version of this article described events that we could not verify. It has been rewritten from verifiable sources, which are linked in the text.

South Korea's deepfake problem is real, and it is not new. In September 2024, as public anger over sexually explicit deepfakes of women and girls reached the streets of Seoul, the National Police Agency reported that 318 people had been arrested for deepfake crimes that year, 251 of them (78.9%) teenagers, and that roughly 60% of the people targeted in deepfake cases investigated from 2021 to 2023 were minors, according to The Korea Times.

The earlier version of this article claimed that Korean criminal law was "powerless" against deepfakes and described a bill said to have been tabled in August 2026. Both claims were wrong. Korea has had a specific criminal offence for sexual deepfakes since 2020, raised the penalties and added possession and viewing offences in 2024, and has banned campaign deepfakes in the run-up to elections under an election-law amendment adopted in December 2023 and in force since January 2024. This article explains those laws, how the courts apply them, what the enforcement data show and where the debate stands, with comparisons to Türkiye and the EU.

1. 2020: a stand-alone deepfake offence

The starting point is Article 14-2 of the Act on Special Cases Concerning the Punishment of Sexual Crimes (성폭력처벌법, the "Sexual Violence Punishment Act"). It was inserted by Act No. 17086 of 24 March 2020 and took effect on 25 June 2020. The official statement of reasons published with the amendment says plainly why: harm from deepfakes that edit a specific person's body into sexual material was increasing, and the existing provisions made such conduct either difficult to punish or punishable only lightly.

The 2020 version was already technology-neutral. It covered anyone who, for the purpose of distribution, edited, synthesised or processed footage, images or audio of a person's face, body or voice, against that person's will, into a form capable of causing sexual desire or shame. The penalty was up to five years' imprisonment or a fine of up to 50 million won. Distributing the result, including distributing material that had been made with consent but was later shared without it, carried the same penalty. Doing so for profit over an information network carried up to seven years.

2. 2024: the law for heavier penalties

The 2024 scandal exposed a different gap: the 2020 offence required an intent to distribute, and simply keeping or watching such material was not a crime. The response came within weeks. The National Assembly's Legislation and Judiciary Committee approved the package on 25 September 2024, The Korea Times reported, and Act No. 20459 was promulgated and took effect on 16 October 2024. In the current text of Article 14-2 and the surrounding provisions:

  • Creation alone is a crime. The words "for the purpose of distribution" were deleted. Editing, synthesising or processing a person's image or voice into sexual material against their will is punishable by up to seven years' imprisonment or a fine of up to 50 million won, the same range as for illegal filming under Article 14.
  • Distribution carries the same seven-year maximum; distribution for profit over an information network now carries a minimum of three years.
  • Possession, purchase, storage and viewing became offences in their own right (Article 14-2(4)): up to three years' imprisonment or a fine of up to 30 million won.
  • Threats and coercion using such material carry at least one year and at least three years respectively (Article 14-3). Habitual offending raises the penalty by up to half (Articles 14-2(5) and 14-3(3)), and attempts are punishable (Article 15).

The same package amended the child protection law: according to The Korea Times, using sexual exploitation material to blackmail a child or teenager now carries at least three years, and coercion at least five.

Two further amendments followed in December 2024. Act No. 20535 (in force June 2025) allowed police to use undisclosed-identity and undercover investigations for digital sex crimes against adults, a power previously limited to cases involving minors. Act No. 20575 (in force 21 June 2025) made confiscation of criminal proceeds mandatory and introduced what is now Article 23-2: once police confirm that reported material is circulating online, they must ask the media standards commission and the relevant service providers or board operators to delete or block it.

The penalties in the earlier version of this article (five to ten years, up to fifteen for minors) did not come from any of these laws. The real maximum for creating or distributing a sexual deepfake of an adult is seven years.

3. How the Supreme Court is applying the law

Possession is a continuing offence. In Case 2026Do541 (30 April 2026), the defendant had made or saved 195 synthetic images between 2019 and 2020, including composites using a university classmate's face, and had obtained illegally filmed material from Telegram groups. He still had the files in December 2024. The appeal court acquitted him of possession on the ground that, after the new offences took effect, he had not been shown to do anything new to maintain or strengthen his control over files acquired earlier. The Supreme Court reversed. "Possession" means keeping material under one's control, so the offence continues for as long as that control lasts. New offences cannot reach conduct before they take effect, but continuing possession after 16 October 2024 can be punished, and no new act of "renewing" possession is required. The case was sent back to the Gwangju High Court.

Deepfakes of real minors. In Case 2024Do17801 (14 August 2025), the Court held that a composite placing a real teenager's face on another body is not material in which "a child or juvenile appears", because what appears is an image created by the maker, not the child. It may still be material featuring "a representation clearly recognisable as a child or juvenile" under the Act on the Protection of Children and Youth against Sex Offences, which carries life imprisonment or at least five years for production. Whether it is depends on an overall assessment of the figure's appearance, the real person's age and identity, how and where the image was made, and its setting. In that case the lower court's conviction under the deepfake offence in Article 14-2 stood.

4. Elections: a 90-day ban since 2024

The earlier version of this article presented a 90-day pre-election ban on deepfakes as a novelty of a 2026 bill. It has been in the Public Official Election Act since an amendment dated 28 December 2023, which took effect on 29 January 2024:

  • Article 82-8(1): from 90 days before an election until election day, no one may make, edit, distribute, screen or post, for campaign purposes, AI-made virtual sound, images or video that are hard to distinguish from reality. Breach carries up to seven years' imprisonment or a fine of 10 to 50 million won (Article 255(5)).
  • Article 82-8(2): outside that window, campaign deepfakes must be labelled as AI-made virtual information in the form set by National Election Commission rules. An unlabelled deepfake can lead to an administrative fine of up to 10 million won (Article 261(3)(4)), and the false-statement offences are aggravated where the falsehood is spread through an unlabelled deepfake (Article 250(4)).

The rule was first applied at the April 2024 general election. Reporting before the June 2025 presidential election, Nocut News noted that no one had been punished under it at the 2024 election and that interpretation and case law were still thin, while deepfakes of leading candidates were already circulating and the Election Commission had set up a special response team on 9 April 2025.

5. Labelling: the AI Basic Act

The "watermark mandate" described in the earlier version also already exists, in a different form. Article 31 of the Framework Act on the Development of Artificial Intelligence and the Establishment of a Foundation for Trust (the AI Basic Act), in force since 22 January 2026, requires AI business operators to give advance notice when a product or service uses generative or high-impact AI, to indicate that outputs were generated by generative AI, and, where a system produces virtual sound, images or video that are hard to distinguish from reality, to notify or label this in a way users can clearly recognise. Artistic and creative works may be labelled in a way that does not hinder their display or enjoyment. This is a transparency duty on businesses, not a criminal offence, and it does not bind the individual who uses a tool to make a deepfake.

6. What the enforcement data show

The numbers available from sources we could verify are limited but instructive:

  • The police figures for 2024 quoted above (318 arrests by September 2024, 78.9% of them teenagers) show who was being caught: mostly young people.
  • A 2023 academic study found that 9,006 deepfake sexual videos were referred to the Korea Communications Standards Commission for review between June 2020 and August 2023, but that deletion was achieved in only 410 cases, 4.5% of the correction requests. The study reports deletions; its abstract does not give a separate figure for access-blocking.

We were not able to verify official nationwide figures for 2025 and 2026 from primary sources, so we do not cite any. The figures we do have suggest that penalties were not the main gap: the harder problems are identifying offenders and getting material taken down quickly. The 2024–2025 amendments on undercover policing and police takedown requests are aimed at exactly that.

7. The policy debate now

Sexual content of people who do not exist. The live legislative question in 2026 is not heavier penalties for deepfakes of real people but whether to extend the criminal law to AI-generated sexual content of fictional people. According to Weekly Kyunghyang, a bill led by Democratic Party lawmaker Heo Young, submitted on 11 September 2025 and placed before the Legislation and Judiciary Committee on 27 July 2026, would add an Article 14-4 punishing the creation or distribution of virtual figures "that can be recognised as real people" in sexual form (up to seven years or 50 million won) and their possession or viewing (up to three years or 30 million won). Its sponsor cited an August 2025 acquittal involving AI nude composites shared in a Telegram chat room. The magazine reports that the court's actual reason was that prosecutors had not proved the woman depicted was a real person, and that the prosecution did not appeal. Critics object that the "recognisable" standard is vague, that the penalties are disproportionate, and that the bill would make fictional adult content a sex crime without any real person being harmed. The bill was in subcommittee review at the time of that report.

Viewing and knowledge. The viewing offence goes well beyond the EU minimum standard discussed below. During the committee stage, lawmakers made clear that people who view or store such material unknowingly are not to be punished (The Korea Times), and in Case 2026Do541 the Supreme Court described possession as beginning when a person knowingly takes control of the material. The boundary between deliberate viewing and incidental exposure in group chats will be tested case by case.

Proof and authenticity. As synthetic media improves, courts face two opposite risks: genuine material dismissed as fake, and fake material accepted as genuine. That is a question for forensic practice and procedure rather than for new offences.

8. Türkiye and the European Union

Türkiye. The Turkish Penal Code (Law No. 5237) has no deepfake-specific offence. Prosecutors have to work with general provisions: violation of the privacy of private life (Article 134: one to three years, doubled where it is committed by recording images or sounds, and two to five years for unlawfully disclosing images or sounds of a person's private life); insult (Article 125: three months to two years or a judicial fine); unlawfully giving, disseminating or obtaining personal data (Article 136: two to four years); and, where children are concerned, Article 226(3), which since 2016 has expressly covered "representative child images" and "persons appearing to be children" (five to ten years for production; two to five years for, among other things, possessing such material). Whether a wholly fabricated image counts as an image "relating to private life" under Article 134(2) is open to argument, and we found no provision that criminalises merely possessing or viewing a sexual deepfake of an adult. On removal, Article 9/A of Law No. 5651 lets a person apply directly for access-blocking on privacy grounds. Since Law No. 7590 of 24 July 2026 the application goes to the Cyber Security Presidency; providers must comply within four hours, and the applicant must put the request before a criminal judge of peace within 24 hours; if the judge does not rule within 48 hours, the measure lapses. The consolidated text shows the general content-removal provision in Article 9 as annulled by the Constitutional Court (decision E.2020/76, K.2023/172). A general AI bill (No. 2/2234), submitted on 24 June 2024, remains in committee according to the parliamentary record.

European Union. Directive (EU) 2024/1385 on combating violence against women requires Member States to criminalise "producing, manipulating or altering and subsequently making accessible to the public" material that makes it appear a person is engaged in sexually explicit activities without consent, where this is likely to cause serious harm (Article 5(1)(b)), with a maximum penalty of at least one year (Article 10(4)). Member States must transpose it by 14 June 2027. The EU floor is therefore narrower than Korean law: it targets making material public and does not require private creation, possession or viewing to be criminalised. On transparency, Article 50(4) of the AI Act has required deployers to disclose deepfakes since 2 August 2026, with a lighter form of disclosure for evidently artistic, satirical or fictional works, according to the European Commission.

Burhan Doğuş Ayparlar's View

In our view, Korea's experience points the other way from the earlier version of this article. The problem was never that the law could not name the wrong. Korea named it in 2020, and by 2024 had built a regime that criminalises creation without distribution, possession and viewing, and threats, with undercover powers and police-initiated takedown requests behind it. Measured by the text of the statute, it is one of the most far-reaching regimes we know of.

The harder lessons are about reach and proportion. A deletion rate of 4.5% for material referred to the regulator between 2020 and 2023, and a perpetrator group dominated by teenagers, suggest that deterrence depends more on detection, school-level prevention and fast removal than on raising maxima again. The current proposal on fictional persons shows where criminalisation runs out of clear justification: once no real person can be identified, the criminal law is protecting something more abstract, and vague standards such as "recognisable as a real person" sit uneasily with legal certainty.

For Türkiye, the comparison is useful in a narrower way. The general offences in the Penal Code fit fabricated imagery poorly, especially for adults. A targeted offence along the lines of Article 5(1)(b) of the EU Directive, covering creation and dissemination of non-consensual sexual synthetic imagery of identifiable persons, would close the most obvious gap without importing the most contested part of the Korean model, the criminalisation of private viewing. Whatever form it takes, it should come with credible forensic capacity, because in deepfake cases both the accusation and the defence will increasingly turn on the question of what is real.

Sources

This article is general information about the law as of 19 October 2026 and is not legal advice. Laws, court decisions and bills referred to may change.