What Naver Filed the Week Seoul Named the Portals

Thirty million won, a shade over twenty thousand US dollars, is the ceiling for shipping an AI-generated advertisement in Korea without the visible label the country's newest AI law now demands. Not the total damages. Not the punitive multiple. Just the plain administrative fine per infraction. It is a small number relative to what a mid-size beauty brand spends on a single influencer week. That is the interesting part.

That number matters because on July 7, a separate law layered a different obligation on top, and on July 8 the Korea Communications Commission named the eight portals and platforms that must, from now, publish half-yearly transparency reports on how they handle false or manipulated information. Four are Korean: Naver, Kakao, Nate, DC Inside. Four are foreign: Google, Meta, TikTok, X. And the same week, Naver pushed its AI Safety Framework to version 2.0, extending its own safety governance from the model outward to the service the user actually sees. If you buy or sell digital advertising in Korea, those three moves are now the room you work inside.

What did July 7 change?

The revised Act on Promotion of Information and Communications Network Utilization and Information Protection, the country's Network Act, was promulgated on January 6, 2026 and its enforcement decree came into force on Tuesday July 7. It applies to any online platform whose average daily user base in Korea has exceeded one million over the past three months. The eight names the KCC published on July 8 are the first cohort. Each must operate a reporting channel for illegal and manipulated information, write and publish its own moderation rules, and file the six-monthly transparency report. Non-compliance carries administrative fines of up to ten million won, a modest sum but a real audit trail.

The one-million-daily-user threshold is not a hypothetical. Naver in a routine week runs well past forty million Korean users on its portal; Kakao's messenger touches nearly every smartphone in the country. The KCC did not have to hunt for names; the list wrote itself. The Commission used its accompanying guidance the same week to insist, on the record, that the law is not a government censorship instrument and that moderation criteria remain with the platform. Whether every civil-society reader believes that phrasing is a separate question. What matters operationally is that the moderation posture is now a document you can be asked to produce.

Where does the AI Basic Act sit in this?

Alongside it, on a different clock. The AI Basic Act took effect on January 22, 2026, with a one-year grace period on administrative fines. For any advertising or media output that is AI-generated and difficult to distinguish from reality, the Act requires a visible label, and per the Ministry of Science and ICT's guidance as summarised by the Stimson Center, a stricter visible mark on content with a high potential to infringe rights, plus an in-metadata watermark on lower-risk creative work. An in-metadata watermark is what it sounds like: machine-readable, invisible to the eye, embedded in the file itself. A visible label is what a viewer can see without opening a properties panel. Both, at different tiers. The fine ceiling is the thirty-million-won figure per infraction.

The two rules are not the same rule. One asks a platform to explain how it moderates false content. The other asks a producer to mark what they made. Anyone reading them as one law is going to trip on the seam.

Who filed a document in return?

Naver did. On July 8, at the Seoul Forum on AI Safety and Security, the company published ASF 2.0, an expansion of the framework it first outlined at the 2024 AI Seoul Summit. The 1.0 version was model-centric: evaluate the model, categorise its risk, cap high-risk deployments. The 2.0 version extends the same discipline out to the service layer, meaning what the user actually clicks, sees and shares. Read the company's own release and you find a taxonomy of service-level risks, an impact-assessment matrix by application domain, and a company-wide execution system called CHEC 2.0. It is the kind of document a legal team writes when it has already read the next law before the next law reads them.

Naver is not a small platform doing a small favour. It is Korea's dominant search, portal and shopping surface, with a domestic depth of penetration Google has not come close to matching here. When Naver publishes a framework, the Ministry of Science and ICT reads it. When Naver publishes a framework the same week the KCC hands out its designation letters, the Ministry reads it as a submitted brief.

The choreography, if you have watched Korean regulatory sequencing before, is familiar. Tuesday July 7: the decree lands. Wednesday July 8: the KCC publishes the eight-name roster and Naver publishes ASF 2.0. The regulator files the rule. The domestic champion files a framework the regulator can point to as a template. The foreign platforms file objections. Everyone waits for the enforcement letters.

What does this do to advertising, in practice?

The interesting friction is not at the platform layer. It is downstream, at the ad-buyer's desk. Korea has been an unusually active market for virtual-model marketing, from Rozy onward, with a long tail of quieter Instagram accounts pushing beauty, health and food products through what look like humans and are not. The AI Basic Act requires those virtual humans, when used in ads, to be labelled visibly where they appear on screen or named in the opening line of a text placement. The Korea Fair Trade Commission has said that fake-doctor and fake-expert endorsements, especially in food and pharmaceutical categories, will be treated as unfair or deceptive labelling. Regulators are trying to bend a real curve: the Food and Drug Safety Ministry flagged more than ninety-six thousand illegal online ads in 2024.

Now overlay the Network Act. If Naver's Shopping tab or Kakao's advertising surface carries a virtual-model ad from a small direct-to-consumer brand and the ad lacks the label, the ad seller is on the hook for the thirty-million-won administrative fine. The platform, as of July 7, is separately on the hook to explain in its transparency report why the ad reached a user. Two regulators, two ledgers, one ad. The Seoul ad-tech people I read call this the double-book problem, and they are right.

Where is the honest disagreement?

Even-handed reading is worth the airtime here. The Information Technology and Innovation Foundation argued in September 2025 that the AI Basic Act sweeps too broadly, defines high-impact AI too loosely, and burdens firms with process-heavy reporting rather than performance-based oversight. The critique is not empty. When only a small minority of Korean AI startups have working compliance systems in place, a labelling regime with a twelve-month grace period is a real transition tax on the small end of the market. The TechPolicy.Press analysis makes the harder claim: being first is not the same as being right, and Korea may find itself owning definitional errors the EU is still busy calibrating.

That critique has weight. My counterargument is narrower than it sounds. The thirty-million-won fine is not what will discipline the Korean ad market. It is the audit trail: the file a platform now has to produce, the labelled-versus-unlabelled column an ad-ops team now has to keep, the six-monthly report a compliance lead now has to file. That paperwork will change behaviour. The fine is the flag on top of the paperwork, not the paperwork itself. If your position is that the paperwork is disproportionate, argue that. But the labelling requirement, for AI outputs difficult to distinguish from reality in a market where a fake-doctor endorsement in a supplement ad is a real scam category, is a reasonable ask that landed clumsily. Those are two different objections, and only the second one is worth spending airtime on.

The seam is visible in Myeongdong

In Myeongdong on a rainy weekday afternoon, a fifty-foot LED billboard cycles between a K-drama trailer, a coffee promotion, and a skincare ad featuring a woman who is, on close inspection of the fine print at the corner of the frame, marked with three small Hangul characters: AI 생성, meaning AI-generated. The mark is legible if you know to look. The operator behind the display sits inside the eight-name KCC roster. The skincare brand sits under the Basic Act. The label is the seam between them, printed at what looks to be twenty-four points on a screen that seats twelve stories of pedestrians below it. It is neither an obstruction nor a revelation. It is the compliance layer, made physical, standing exactly where the two Korean laws meet.

Tarry Singh is the founder and CEO of Real AI, an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan for Energy AI, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.