South Korea Moves to Criminalise AI Sexual Content of People Who Don't Exist

What happened

The bill, sponsored by Democratic Party lawmaker Huh Young and nine colleagues, would amend the Act on Special Cases Concerning the Punishment of Sexual Crimes, one of South Korea's core statutes on sexual offences. As reported by The Korea Times, it was referred to the Legislation and Judiciary Committee's subcommittee in August 2026 and is under detailed review.

Its innovation can be put in a single sentence: AI-generated sexually explicit material becomes a matter for the criminal law even where there is no identifiable real victim. That goes beyond the logic on which deepfake regulation has rested so far — that a real person's identity and dignity have been violated.

What the bill criminalises, and the penalties

According to press reports, the bill grades sanctions in two tiers according to the gravity of the conduct:

Creation and distribution: up to seven years' imprisonment or a fine of up to 50 million won for creating or distributing AI-generated sexually explicit material depicting fictional or non-existent people.
Possession, purchase and viewing: up to three years' imprisonment or a fine of up to 30 million won for possessing, buying, storing or viewing such material.

It is notable that these ranges mirror the sanctions South Korea introduced in 2024 for deepfake sexual content of real people. The legislature's message is explicit: whether the person depicted actually exists is not decisive to how socially harmful the conduct is.

Why now? An acquittal and exploding statistics

Two grounds stand out in the bill's rationale. The first is a judicial gap. In 2025, a court in Goyang acquitted a defendant who had distributed AI-generated nude images because there was insufficient evidence that the images belonged to a real victim. Because existing law turns on a "real person" element, the criminal law could not engage once a synthetic face or body was used.

The second is scale. According to figures cited by the bill's sponsors, digital sex crimes in South Korea rose by 66 per cent between 2015 and 2024, while deepfake sex-crime cases climbed from 31 in 2020 to 550 in 2024. With generative models now ubiquitous, producing photorealistic sexual imagery — without even needing a real person's photograph — has become a matter of seconds.

The picture continues the public shock of the 2024 Telegram deepfake scandal, when mass sharing of deepfakes targeting students and young women came to light and the country rapidly enacted one of the world's toughest regimes, criminalising not only the creation but also the possession and viewing of deepfake sexual content. We covered that process in our report on the day South Korea's Deepfake Sexual Violence Act passed parliament and in our analysis of South Korea's deepfake crisis.

The existing framework: real people and children are already covered

To understand what the bill really changes, it helps to map South Korea's current legal architecture:

Deepfakes of real people: the 2024 amendments imposed heavy penalties for creating sexually explicit deepfakes of real people without consent, and made possessing, purchasing, storing or viewing such content an offence punishable by up to three years' imprisonment or a fine of up to 30 million won.
Children and young people: South Korea's law protecting children and youth from sexual abuse already covers material featuring people or representations clearly recognisable as children or young people. Fictional depictions that appear to be minors are therefore already criminal.
Obscene content: the Criminal Act and legislation on information and communications networks generally prohibit the distribution of obscene material. But those provisions focus on distribution; possessing or viewing adult-appearing fictional content is, as a rule, not an offence.
The AI Basic Act: in force since 22 January 2026, it imposes labelling and notice duties for deepfake outputs that are hard to distinguish from reality — a transparency tool, not a criminal provision.

Against that backdrop, the bill's real effect is to move adult-appearing synthetic sexual content that is not based on a real person into the category of "sexual crime," and to extend sanctions all the way to possession and viewing.

The legal debate: can there be a crime without a victim?

The bill revives one of criminal law's oldest questions for the age of AI: on what basis can conduct with no concrete victim be punished?

The case for

Proof of a real person is becoming impossible. Generative models can blend real faces beyond recognition, producing content that "belongs to no one but resembles someone." A system that depends on proving the victim's identity lets an offender escape liability simply by altering a face slightly.
Normalisation and social harm. The mass production and consumption of non-consensual sexualisation creates an ecosystem that normalises digital sexual violence, particularly against women.
Breaking the demand chain. Punishing possession and viewing aims to deter production by targeting the demand side of the deepfake market.

The case against

Freedom of expression. South Korea's Constitution guarantees free expression, and its Constitutional Court has accepted that obscene expression cannot simply be placed outside constitutional protection. Criminalising the possession of fictional adult content without a concrete victim raises serious questions of proportionality.
Legal certainty. Where is the line between "AI-generated" and a hand-drawn illustration? Is a drawing retouched with an AI-assisted editing tool in scope? How will stylised content such as anime and webtoons be treated? Criminal norms must be clear and foreseeable, yet these questions remain open.
Criminalising viewing. Making mere viewing an offence may, in practice, invite inspection of personal devices and intrusions on privacy.
Enforceability. How to prove technically that content is synthetic and not based on a real person creates evidential problems in the opposite direction as well.

At the heart of the debate lies the question of what legal interest the criminal law protects: an individual's dignity and identity, or a social order grounded in respect for sexual autonomy? South Korea's legislators propose extending toward the latter; critics warn that this risks turning criminal law into moral policing.

How other legal systems treat fictional sexual material

South Korea is not the first country to ask whether the criminal law can reach sexual imagery that depicts no real person. Comparative law offers three broad models, and the Korean bill sits at the most expansive end:

The harm-to-real-persons model (United States). In Ashcroft v. Free Speech Coalition (2002), the US Supreme Court struck down a federal ban on "virtual" child sexual imagery that involved no real children, holding that speech cannot be banned merely because it is thought to encourage harmful conduct. Congress responded with the PROTECT Act of 2003, which reaches such material mainly where it is legally obscene. American law therefore protects real people strongly but treats purely fictional content with caution.
The realistic-depiction model (United Kingdom). English law criminalises the possession of "extreme pornographic images" that realistically depict certain grave acts, and separately the possession of non-photographic prohibited images of children, such as drawings and computer-generated images. The UK has also moved to criminalise the creation of sexually explicit deepfakes of real adults. The yardstick is the nature and realism of the depiction rather than the existence of a victim — but it is confined to narrowly defined categories.
The minimal-intervention model (Japan). When Japan criminalised the simple possession of child sexual abuse material in 2014, lawmakers expressly left manga, anime and computer graphics outside the scope after an intense debate on freedom of expression.

Against this spectrum, the Korean bill is remarkable in two respects: it targets adult-appearing content, and it attaches liability to possession and viewing across the board rather than to narrowly defined "extreme" categories. That is why constitutional scrutiny is likely to be intense.

The evidence problem: proving that content is synthetic and victimless

Criminal liability is only as effective as the evidence behind it. The bill shifts the evidential question rather than eliminating it. Under current law, prosecutors struggle to prove that an image depicts a real person; under the bill, a defendant accused of an offence concerning a real person may argue the content is purely synthetic, and prosecutors may need to show the material is AI-generated at all.

Several technical tools are relevant. Content-provenance standards such as C2PA can attach cryptographically signed metadata recording how a file was created, and invisible watermarks can mark AI outputs at the moment of generation. South Korea's AI Basic Act already pushes providers toward labelling deepfake outputs. But metadata can be stripped, watermarks degraded by compression or editing, and detection classifiers produce false positives and false negatives. We examined the strengths and limits of these tools in our analysis of mandatory AI watermarking. In practice, criminal cases will likely rest on a combination of device forensics, platform logs and the defendant's own communications rather than on any single technical marker.

The legislative road ahead in Seoul

Under South Korea's legislative procedure, a bill reviewed in committee subcommittee must be approved by the full committee, pass the Legislation and Judiciary Committee's review of legal consistency and wording, and then be adopted in a plenary session before it is sent to the government for promulgation. Because this bill amends a statute within the Legislation and Judiciary Committee's own remit, the same committee handles both the substantive review and the wording review, which can speed things up. Bills that are not enacted by the end of the National Assembly's term lapse automatically, so the political will of the governing and opposition parties will decide the timing. In the 2024 deepfake reforms, public pressure led to enactment within weeks; whether this more controversial bill will see the same momentum is an open question.

Consequences for platforms and AI developers

If the bill becomes law, its effects will not stop at individual users:

Generation tools. Image-generation services may have to tighten the safety filters that block sexual outputs for the Korean market; a defence that "the user generated it" will be weighed alongside the question of whether the tool was designed in a way that facilitates such content.
Hosting and distribution. Platforms' notice-and-takedown processes may have to operate for content not based on any real person, dramatically expanding the scale of content moderation.
The virtual-character economy. South Korea's fast-growing virtual-idol and digital-character sector will look for clear lines on adult content. We have discussed the legal dimensions of that economy in K-pop's synthetic revolution and in whether an entity made of code can have a right of publicity.

A compliance checklist for AI services reaching Korean users

Even before the bill's fate is known, companies offering image-generation, chatbot or hosting services to users in South Korea can reduce their exposure with steps that are good practice under any version of the law:

Model-level safeguards. Filter sexually explicit outputs by default, block prompts that seek sexualised depictions of real people or minors, and red-team these filters regularly against evasion techniques.
Age and identity assurance. Where adult content is permitted at all, put proportionate age-assurance measures in place and prevent the upload of third-party photos as generation references.
Provenance by design. Label and watermark generated images in line with the AI Basic Act's transparency duties, so outputs remain traceable.
Takedown readiness. Set response times for reports of synthetic sexual content and establish a working channel with the Korea Communications Standards Commission and law-enforcement authorities.
Log retention and privacy. Balance the evidential value of generation logs against data-minimisation duties under Korea's Personal Information Protection Act.
Transparency reporting. Publish how many reports were received, how fast content was removed and how filters performed.

The global context: the world protects real people — Korea goes a step further

The global wave of regulation aimed at deepfake sexual content focuses overwhelmingly on protecting real, identifiable people:

United States: the TAKE IT DOWN Act, enacted in 2025, criminalised publishing non-consensual intimate images of identifiable people — digital forgeries included — and imposed rapid takedown duties on platforms.
European Union: the directive on combating violence against women and domestic violence requires member states to criminalise the non-consensual production or manipulation of intimate material, including deepfakes.
California: SB 1276, part of the package signed in September 2026, brought AI-generated material within child-abuse offences; we covered it in our report on California's "Adam's Law".

The comparison reveals what is distinctive about the Korean bill. A "fictional depiction" approach to child protection is widely accepted, but punishing the possession and viewing of adult-appearing content not based on any real person is a rare choice among democracies.

What it means for Turkey

In Turkish law, deepfake content involving real people can be assessed under criminal provisions on privacy and personal data, the offence of obscenity, civil-law rules protecting personality rights, and the data-protection law. The access-blocking and removal mechanisms of Law No. 5651 on internet content may also apply. There is, however, no provision specific to synthetic sexual content that depicts no real person.

The Korean debate leaves two important questions for Turkey. First, whether an approach built on proving a real person is adequate against generative models that blend faces. Second, how any future rule would balance free expression, legal certainty and proportionality. In our view, the sounder first step for Turkey would be a rule that expressly defines non-consensual synthetic content of real people, imposes rapid takedown duties on platforms and eases the victim's burden of proof. We discussed how such a framework might be built in our 100-article draft proposal for a Turkish AI Act.

What to watch next

How far the bill changes at subcommittee stage will be decisive. The definition of "AI-generated," the scope of the viewing offence, and possible exceptions for artistic or stylised content deserve close attention. If enacted, a constitutional challenge on free-expression and proportionality grounds would not be surprising. Internationally, South Korea's move may push the "victimless synthetic content" debate onto other countries' agendas.

Frequently asked questions

Has the bill become law?

No. According to press reports it was referred to the Legislation and Judiciary Committee's subcommittee in August 2026 and is under detailed review. The text may change during that process.

Aren't fictional depictions of children already illegal?

Yes. South Korea's law protecting children and youth already covers representations clearly recognisable as minors. The bill's real novelty is that it targets adult-appearing content not based on a real person.

Would anime or drawings be covered?

The bill targets AI-generated content, but where drawings created or edited with AI-assisted tools fall is one of the most uncertain points of the debate. The definitions in the final text will decide it.

Are foreign platforms affected?

AI and hosting services that serve users in South Korea may, in practice, have to adapt their filters and takedown processes to comply.

Why criminalise viewing and not just distribution?

The sponsors argue that punishing the demand side is necessary to deter production. Critics respond that punishing mere viewing of content with no concrete victim is disproportionate and invites intrusions on privacy.

Is there a comparable rule in Turkey?

Deepfake content affecting real people can be addressed under existing criminal, civil and data-protection provisions, but there is no rule specific to synthetic sexual content that depicts no real person.

Doesn't the AI Basic Act already deal with deepfakes?

Only in part. The AI Basic Act requires notice and labelling for AI outputs that are hard to tell apart from reality, but it is a transparency law, not a criminal statute. It does not make the creation or possession of synthetic sexual content an offence.

What would happen to content people already have stored?

As a general principle of criminal law, conduct is judged under the law in force when it takes place. But because possession is a continuing state, keeping such material after a new law enters into force could itself fall within its scope. The transitional provisions of any final text will matter.

Expert Opinion

This section reflects my personal assessment as the founder of this site and an AI ethics & compliance counsel.

In my view, the Korean bill correctly diagnoses a real problem but, in part of its remedy, strains the limits of criminal law. The diagnosis is right: generative models can defeat a protective architecture that depends on proving a "real person." If an offender who blends several faces into an image that "belongs to no one" can escape liability under current rules, that gap must be closed. The Goyang acquittal shows the gap is concrete, not theoretical.

But punishing creation and distribution is not the same as punishing the mere possession or viewing of adult-appearing content with no concrete victim. The latter carries serious risks under the principles of legal certainty and proportionality, and threatens to shift criminal law from a harm-based system toward moral policing. A more balanced model would target the tools and distribution chains that facilitate non-consensual sexualisation, while easing the burden of proof in the victim's favour and imposing effective takedown duties on platforms.

My lesson for Turkey is this: deepfake regulation should be built not on an abstract binary between "real" and "fictional" people, but on the question of where concrete harm arises. Our priority should be fast, effective, victim-centred protection against non-consensual synthetic images of real people. In the age of AI, criminal law should neither surrender to technology nor try to solve every new risk by punishment.

This article is for information only and does not constitute legal advice. Facts about the bill are based on publicly available sources, principally The Korea Times report of 11 August 2026; the text may change during the legislative process. The analysis and assessments are the author's own.