What happened

On 30 September 2026 the Korea Media and Communications Commission (방송미디어통신위원회, KMCC) adopted its "first-term vision and major policy measures" at its 39th plenary meeting of 2026, held at the Government Complex in Gwacheon. The vision carries the title "a basic media society together with the people" (국민과 함께하는 미디어 기본사회). It has four goals and twelve policy tasks, and is the first vision since the Commission was launched in October 2025.

According to the Commission's press release, the goals are universal access, active public participation, autonomous choice founded on safety and trust, and a sustainable ecosystem. The AI material sits under the third goal, in general terms: the Commission will reorganise the "distribution and management system" for AI-generated content to strengthen users' right to know and to choose, raise the transparency of recommendation algorithms, and prepare AI ethics principles for broadcasting and media. The release itself gives no legislative detail and no target dates. The specifics, which are considerably more ambitious, were given at the briefing and reported by Electronic Times, MTN, Nocut News and others.

A note on the body itself. The KMCC is not simply a renamed Korea Communications Commission. The Act on the Establishment and Operation of the Korea Media and Communications Commission, passed by the National Assembly on 27 September 2025, abolished the Korea Communications Commission and created the KMCC as a central administrative agency under the President; the new body was launched on 1 October 2025. It also took over pay-TV and new-media policy from the Ministry of Science and ICT (MSIT). The Act provides for seven commissioners. Its first chair, Kim Jong-cheol, a constitutional law professor at Yonsei University, was nominated in late November 2025 and took office in December. He described the vision as "not a completion but a beginning".

What the plan says, item by item

A labelling duty for people who post AI content. The Commission will seek an amendment to the Act on Promotion of Information and Communications Network Utilization and Information Protection (the Network Act) requiring users who post AI-generated content to indicate that it was made with AI. Ordinary users would also be prohibited from damaging, forging or altering an AI label.

A management duty for portals and platforms. Portal and platform operators would be made responsible for managing whether posters and users comply with the labelling duty. The press reports describe this as a responsibility to "manage" compliance, not an obligation to detect every unlabelled item, but the drafting has not been published.

Timing and alignment. The release gives no dates. According to MTN and The Fact, the Commission aims to pursue the labelling amendment by December 2026, and MTN reports that it will consult MSIT so that the amendment is consistent with the AI Basic Act. The Fact adds that, for a wider AI user-protection framework, both a separate statute and a new "AI user protection" section in the Network Act are under review as medium- to long-term options.

Recommendation algorithms. Platforms would have to disclose, in their terms of service and transparency reports, whether recommendations are applied, how they work and any major changes. Users would be able to choose whether to receive algorithmic recommendations and in what form. According to MTN, Electronic Times and The Fact, the related Network Act amendment is targeted for June 2027.

AI ethics principles for broadcasting. According to MTN and Nocut News, by the end of 2026 the Commission intends to issue "AI ethics principles for the broadcasting sector". Broadcasters using AI would keep final human verification and responsibility, and AI-generated content would be flagged on screen or by audio. Stricter, separate principles for news and current affairs would discourage AI-generated "evidence" that could be mistaken for real footage or recordings, and the recreation of real people. A Commission official told Nocut News that the principles will be a guideline without legal force, aimed at voluntary implementation and finalised after consultation.

The rest of the package. It also includes a unified media law covering broadcasting and OTT services (according to Nocut News, a draft is due in December 2026), a renewable connecting-information identifier (CI) after repeated data leaks, and, according to Financial News, a transparency centre to support private fact-checking.

The labelling idea is not new. On 10 December 2025 the government announced an "AI-generated content labelling system" against fake AI doctors in advertising, promising a Network Act amendment in the first quarter, or first half, of 2026. That date has passed; the new vision widens the plan and resets the target.

What is already law in Korea

This is not Korea's first AI labelling rule. Korea already has two statutory labelling regimes and has legislated a sector-specific advertising ban.

The AI Basic Act, Article 31. The Framework Act on the Development of Artificial Intelligence and the Establishment of a Foundation for Trust (enacted as Act No. 20676 and since amended, most recently by Act No. 21311 of 20 January 2026) has been in force since 22 January 2026. Article 31 sets out three transparency duties:

  • Article 31(1): an AI business operator providing a product or service that uses high-impact or generative AI must notify users in advance that it is based on AI.
  • Article 31(2): an operator providing generative AI, or a product or service using it, must indicate that the output was generated by generative AI.
  • Article 31(3): where an AI system produces virtual sound, images or video that are hard to distinguish from reality, the operator must notify or label this in a way users can clearly recognise. For artistic or creative works, this may be done in a way that does not hinder display or enjoyment.

The duty-bearer is the "AI business operator" defined in Article 2(7): an AI developer, or a business that uses another developer's AI to provide AI products or services. A "user" under Article 2(8) is someone who receives an AI product or service. A person who uses a generator to make a video and posts it is, on that structure, a user and not a duty-bearer, and government guidance reported at entry into force confirms that people using AI merely as a tool for work or creative production are outside the obligation. The same reports say outputs leaving the service should carry a visible or audible watermark or machine-readable metadata, and deepfakes a human-perceivable label wherever they appear.

Enforcement is lighter than the wording suggests. The fine of up to KRW 30 million in Article 43(1) attaches directly only to failure to give the Article 31(1) advance notice. Breaches of Article 31(2) and (3) go through MSIT's fact-finding and corrective orders under Article 40, and a fine follows only if an order is ignored. The government has promised a grace period of at least one year before fines. Under Article 4(1), the Act reaches conduct abroad that affects the Korean market or users.

The Public Official Election Act, Article 82-8. Since 28 December 2023 the election law has contained a deepfake rule that applies to "anyone", not just businesses. From 90 days before election day until election day, campaign deepfakes are banned outright, on pain of up to seven years' imprisonment or a fine of KRW 10–50 million (Article 255(5)). Outside that window they must carry a label, in the form set by National Election Commission rules, stating that the content is virtual information made with AI; failure carries an administrative fine of up to KRW 10 million (Article 261(3)(4)). Unlabelled deepfakes used to spread false statements about candidates attract aggravated penalties (Article 250(4)).

Advertising. On 23 April 2026 the National Assembly passed amendments to the Pharmaceutical Affairs Act, the Act on Labelling and Advertising of Foods and the Cosmetics Act. They ban advertising in which AI-generated doctors, dentists or other professionals vouch for the efficacy of food, medicines or cosmetics. This is a prohibition, not a labelling rule; it is limited to those sectors and takes effect only some months after promulgation. Separately, the Fair Trade Commission's revised endorsement guidelines are reported to require advertisers to make clear when an endorser is an AI-generated virtual person.

The result is a patchwork. AI providers must label at the point of generation. Everyone must label campaign deepfakes. Outside elections and rules aimed at advertisers, nobody is yet required to label AI content when they post it, and no platform is required to check.

What would require legislation, and the hard questions

Every binding element of the plan (the poster's duty, the tamper ban, the platform duty and the algorithm disclosures) needs an Act of the National Assembly; only the broadcast principles, as guidance, do not. A related bill is already pending: in January 2026 Representative Seo Young-seok (Democratic Party) introduced a Network Act amendment that would introduce labelling of AI-generated audio, images and video, prohibit damaging or forging labels, and require platforms to take technical and administrative measures. The Commission's December text, when it appears, will probably either build on that bill or replace it.

The plan is a deliberate move from a supply-side duty to a distribution-chain duty. Article 31 regulates the tool; the KMCC plan would regulate the person who publishes the output and the intermediary that carries it. That shift raises questions the briefings did not answer.

Knowledge. A poster can label only what they know is AI-generated. Content is reposted, screenshotted and recompressed, and many users do not know the origin of what they share. Without a knowledge or intent standard, the duty would be either unenforceable or unfair, and the constitutional protection of expression will force the legislature to choose.

Definitions. The EU AI Act exempts AI that performs "an assistive function for standard editing". A Network Act duty covering any "AI-generated content" would catch filters, upscaling and AI-assisted editing unless drawn carefully. Article 5(2) of the AI Basic Act requires other laws on AI to conform to its purposes, which gives MSIT's consultation real weight.

The platform standard. "Managing" compliance can mean an upload toggle and a policy, notice-and-action on reports, or proactive detection. The last would sit uneasily with platform liability principles and with current detection technology. The framing will decide the cost.

The tamper ban. If the AI Basic Act requires providers to attach watermarks or metadata, a rule against stripping them protects that investment downstream. But metadata can also be lost through ordinary upload processing such as compression, resizing and format conversion, with no one intending to remove it. A tamper ban aimed at users may, in practice, become a design requirement for upload pipelines.

Sanctions. The briefings did not say what would happen to a user who fails to label. Fining individuals for undisclosed AI images would be a significant step.

How the EU compares

Article 50 of the EU AI Act has applied since 2 August 2026. It divides the work between two actors:

  • Article 50(2): providers of generative AI systems must ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. The duty does not apply to the extent that a system performs an assistive function for standard editing or does not substantially alter the input data.
  • Article 50(4): deployers of a system that generates deepfakes must disclose that the content is artificially generated or manipulated, with a lighter regime for evidently artistic, creative, satirical or fictional work. The same applies to AI-generated text published to inform the public on matters of public interest, unless it has undergone human review or editorial control and someone holds editorial responsibility.

The decisive difference is in the definition. Under Article 3(4) a "deployer" excludes a person using an AI system "in the course of a personal non-professional activity". The EU's labelling duty therefore stops at professional users. An individual posting a synthetic image for fun has no Article 50 obligation. The Korean plan would put exactly that individual under a statutory duty.

According to the Commission's Article 50 FAQ, providers of systems placed on the market before 2 August 2026 have until 2 December 2026 to meet the Article 50(2) marking duty. Fines can reach EUR 15 million or 3% of worldwide turnover. A voluntary code of practice on marking and labelling has also been finalised; the FAQ says the Commission and the AI Board have assessed it as adequate.

The EU's platform-side rules sit in a different instrument. The Digital Services Act already requires online platforms to set out the main parameters of their recommender systems in their terms and conditions (Article 27), and requires very large platforms to offer at least one recommender option not based on profiling (Article 38). The KMCC's algorithm plan closely resembles these provisions. For labelling, the DSA lists "prominent markings" for realistic generated or manipulated content as a possible risk-mitigation measure for very large platforms (Article 35(1)(k)). That is a risk-based option for the largest services, not a general duty on all platforms to manage users' labelling.

Korea would not be the first to go this far. China's Measures for Labelling AI-Generated Synthetic Content, in force since 1 September 2025, already require users who publish AI-generated content to declare it and use the platform's labelling tools, prohibit removing or tampering with labels (Article 10), and require platforms to check for embedded labels and flag suspected synthetic content (Article 6). The three elements of the KMCC plan map closely onto that model.

Put simply, the EU labels at the source and asks the largest platforms to manage systemic risk. Korea would add a duty on every poster and a management duty on every platform, which would put its regime, alongside China's, among the most demanding of any major jurisdiction.

Türkiye: no labelling rule, but a bill with a six-hour clock

Türkiye has no AI labelling obligation in force, no horizontal AI law, and nothing in Law No. 5651 (internet) or Law No. 6698 (data protection) requiring AI content to be marked. Harmful deepfakes are dealt with under the general framework: access-blocking and content-removal routes under Law No. 5651, personality-rights claims, and the Turkish Penal Code's general offences.

Two parliamentary initiatives are relevant, and neither has been enacted. A general AI bill (No. 2/2234) submitted on 24 June 2024 has been in committee ever since, according to the parliamentary record. A separate bill by an MHP deputy, reported as amending the Penal Code (Law No. 5237) and other laws, was submitted in autumn 2025; press reports give different dates (early September and 7 November 2025). According to published summaries, it would require AI-generated content that violates personality rights or threatens public security to be removed or blocked within six hours, require deepfakes to carry the statement "Yapay zekâ tarafından üretilmiştir" ("Generated by artificial intelligence"), and hold content providers and AI developers responsible. Reported sanctions include administrative fines of TRY 500,000 to 5 million for a missing label and up to TRY 10 million for failing to remove content. We have not been able to locate the bill in the parliamentary record or confirm its number and status.

The Turkish bill is closer to the Korean plan than to the EU model: it places obligations on content providers, shares the Korean plan's open questions on knowledge, definitions and platform standards, and adds a six-hour clock that Korea has not proposed.

What platforms and content creators serving Korea should do

Work out your role under Article 31 now. A business providing generative AI to Korean users is an AI business operator already under the labelling duty, with or without a Korean office. A service that only hosts user content is not, but would acquire a management duty under the plan.

Preserve labels through the pipeline. If uploads strip metadata or crop visible watermarks, that is the first thing a tamper ban will expose. Platforms should review whether their processing removes provenance information such as C2PA manifests, and decide whether to preserve, display or re-attach it.

Build the poster-side toggle before it is mandatory. An upload-time declaration ("this content is AI-generated"), a visible label and a policy for mislabelling are the most likely core of any "management" duty. Where such features already exist, the question is whether they will meet a Korean statutory standard once one is written.

Treat the election calendar as a hard constraint. The 90-day deepfake ban in the election law already applies to everyone. Producers of political content for Korea need an election calendar and a labelling process for outside the window.

Prepare recommender disclosures. Services already subject to DSA Article 27 have most of the material; others should start with a plain description of main parameters and an opt-out design.

Creators, agencies and broadcasters. Individual creators are not bound by the AI Basic Act today, but election rules bind everyone and the April advertising amendments target AI-generated professional endorsements for food, medicines and cosmetics. Labelling now costs little and avoids a retrofit. Newsrooms using synthetic reconstructions should document human verification, since the December principles, though guidance, will probably become the benchmark for complaints.

What to watch next

  • The labelling amendment text (December 2026 target): its knowledge standard, sanctions on individuals, and the scope of the platform duty.
  • The broadcast AI ethics principles (December 2026) and the algorithm amendment (June 2027).
  • The end of the AI Basic Act's grace period on fines, not before late January 2027 on the government's stated minimum.
  • 2 December 2026 in the EU, when the Article 50(2) grace period ends for systems already on the market.
  • Whether either Turkish AI bill moves out of committee.

Frequently asked questions

Do individual users in Korea have to label AI content today?

Not as a general rule. The AI Basic Act binds AI business operators, not people who use AI tools to make content. The exception is election campaigning: under Article 82-8 of the Public Official Election Act, anyone must label campaign deepfakes, and deepfakes are banned altogether in the 90 days before an election.

Is the KMCC plan binding?

No. It is a policy vision adopted by the Commission. The poster's duty, the tamper ban, the platform duty and the algorithm rules all require amendments passed by the National Assembly. The broadcast AI ethics principles will be a non-binding guideline.

Does Article 31 apply to foreign AI providers?

Yes, where their conduct affects the Korean market or Korean users (Article 4(1)). Operators with no address or place of business in Korea that meet the user-number or revenue thresholds set by Presidential Decree must also designate a domestic agent (Article 36), and failure to do so carries a fine under Article 43.

Is there anything comparable in Türkiye?

Not in force. A bill submitted by an MHP deputy in autumn 2025 would, according to press summaries, require a "Generated by Artificial Intelligence" statement on deepfakes and six-hour blocking of certain AI content, but it has not been enacted and its number and status could not be confirmed.

Burhan Doğuş Ayparlar's View

This section sets out the personal assessment of Burhan Doğuş Ayparlar, the founder of this site and an attorney at law (Türkiye).

The Korean plan addresses a real gap. Labelling at the point of generation works only while the label survives. Once a synthetic video has been downloaded, cropped, re-encoded and reposted, the provider's watermark is often gone, and the person who finally publishes it to a large audience has no obligation at all. The AI Basic Act regulates the factory and leaves the market alone; the KMCC is right that this is the weak point.

Whether the chosen remedy fits the problem is less clear. A labelling duty on every user is easy to announce and hard to design. Most people who share misleading synthetic content do not know it is synthetic; those who do know and intend to deceive are unlikely to comply. A duty without a knowledge standard would catch the innocent and miss the guilty; a duty with one adds little to the fraud, defamation and election offences that already exist. In our view the tamper ban and the platform duty matter more. The first protects provenance information that already exists; the second can be met through interface design (an upload toggle, label display, metadata preservation) without asking anyone to read users' minds. The algorithm proposal, a modest obligation the EU has already tested, is the least controversial part and arguably the most useful.

For Türkiye, the lesson is about sequence. Korea is now trying to add distribution-chain duties on top of a supply-side labelling law it already has. The 2025 Turkish bill would begin with the distribution chain and a six-hour clock, without the provider-side marking that makes labels technically durable. If Türkiye legislates, provider-side machine-readable marking should in our view come first, then a platform duty to preserve and display it, with any user-level obligation confined to knowing deception. That order follows the technology and makes the law more workable.

This article is for information only and does not constitute legal advice. It is based on the Korea Media and Communications Commission's press release of 30 September 2026 and press reports of the same day on the briefing (Electronic Times, MTN, Nocut News, The Fact, Financial News and others). The release states the measures only in general terms; the detailed measures and target dates come from those reports, not from the release or any legislative text; one outlet numbered the meeting the 38th, while the release and most reports give the 39th. Statutory references are to the current Korean texts on the National Law Information Center, Regulations (EU) 2024/1689 and 2022/2065, the European Commission's Article 50 FAQ and an unofficial English translation of China's labelling measures. Information on the 2025 MHP bill comes from press summaries, which differ on its submission date and fines; its number and status could not be confirmed. Statements about Turkish law are general in nature. The analysis and assessments are the author's own.