South Korea's media regulator reviewed more than 17,000 sexual deepfake items in the first eight months of 2026. That already exceeds its entire 2025 total, according to Korean press reports dated 17 October 2026 that the OECD AI Incidents Monitor summarises. The reports followed an October parliamentary audit at which the commission was pressed to speed up the removal of illegal content hosted overseas. We could not independently confirm the exact item counts or the chair's remarks from the primary release, so we use the OECD's rounded figure here.
The case for tougher law is real
The strongest argument for aggressive intervention is that the harm is severe and growing. A report from the Korean Women's Development Institute, cited by the Korea Times, found that reported deepfake sex crimes rose from 31 cases in 2020 to 550 in 2024. Victims are overwhelmingly women, and often minors. Once an image is online, the damage is hard to undo. A legislature that sees this curve and responds with criminal penalties is acting on evidence, not panic.
What Korea has already done
The criminal framework is already substantial. Before 2024, Article 14-2 of the Act on Special Cases Concerning the Punishment of Sexual Crimes covered only producing and distributing false sexual videos. The pre-2024 text set up to five years' imprisonment or a fine of up to 50 million won for each. It had no offence for possession or viewing.
The current text, amended 16 October 2024, raises the maximum for production and distribution to seven years. It adds a new paragraph punishing anyone who "possesses or has seen or purchased" such material with up to three years or a fine of up to 30 million won. Parliament also built in a safeguard. As Yonhap reported through the Korea Times, lawmakers from both main parties added a clause so that people who "unknowingly" stored or viewed such material would not be punished. That mens rea limit is sound drafting. It stops a criminal offence from reaching people who were sent content they did not ask for.
So the 2026 surge did not occur in a legal vacuum. Production, distribution, possession and viewing are all already crimes. More cases reaching review may partly reflect better reporting and detection, not only more abuse. That is plausible, but the data released so far cannot separate the two.
The bottleneck is enforcement across borders
If the law is broad, the problem is reach. Most of the offending content sits on foreign sites and services that Korean courts and platform rules struggle to touch. The regulator's practical lever is access blocking. On 24 August 2026 it said it had tracked down and banned access to 64 foreign websites that let users create pornographic deepfakes with AI, according to the Korea Herald.
That is the right category of action. It targets services built to produce non-consensual sexual imagery of real people. It does not regulate general-purpose AI or ordinary speech. It also works against the actors that matter most, because the offenders and their hosts are largely beyond domestic prosecution. The commission's stated priority of faster takedowns and blocks for overseas-hosted content is therefore where public effort should go.
Where proportionality starts to strain
The live legislative risk is the pending bill from Rep. Huh Young and nine other Democratic Party lawmakers. The Korea Times reports that it would extend Article 14-2 to AI-generated sexual material depicting fictional people, with no real victim. Creating or distributing it would carry up to seven years, and possessing or watching it up to three. The sponsors' stated motive is a Goyang district court acquittal in which, per the same report, the court said the evidence "did not establish whether there were any real victims."
The acquittal exposes a genuine evidentiary gap. Prosecutors cannot always prove that a synthetic face belongs to a real person. But that is a problem of proof, and it can be addressed with narrower tools: presumptions about realistic depictions, or duties on tools that generate recognisable likenesses. Criminalising private possession of wholly fictional imagery is a different step. It removes the identifiable victim that justifies the existing offences. It would also reach creative and private expression, and it would push enforcement agencies toward policing content instead of protecting people. A regulator already stretched across 17,000 items in eight months gains nothing from a much larger category of material to review.
What a proportionate response looks like
- Prioritise speed on real-person content. Measure and publish median time-to-removal for reviewed items, so lawmakers can see whether the chair's pledge is being met.
- Expand cooperation with hosting and payment intermediaries. Blocking 64 sites is meaningful, but sites reappear under new domains. Work with infrastructure providers and cross-border partners.
- Fund review capacity. Press reports in 2025 already flagged that monitoring staff had stagnated while caseloads rose. Speed depends on people.
- Hold the line on identifiable harm. Keep criminal liability tied to real, identifiable victims and to knowing conduct, as the 2024 amendment did.
- Report outcomes. Publish how many reviewed items led to takedown, blocking or prosecution, so rising counts can be read as either detection or abuse.
South Korea has already written the criminal law. The 2026 numbers say the system is overloaded, not under-criminalised. The best answer is a faster, better-resourced and more transparent takedown system aimed at real victims. Extending criminal law to fictional material is not.