Skip to main content
CountryInitial profile

AI Governance Profile

South Korea AI Governance Profile

Initial Observatory profile of South Korea AI governance, centered on the binding AI Basic Act, transparency and high-impact AI duties, advanced-model safety, and the public-sector AI/data administration statutory layer.

Last reviewed
South Korea flag

Profile at a glance

Executive overview

Key signals for South Korea AI governance, structured for quick institutional review before the full profile analysis below.

Region

Asia-Pacific

Governance model

Comprehensive AI law

Regulatory maturity

Advanced

Enterprise impact

High

Public-sector readiness

Advanced

Enforcement maturity

Developing

Enterprise implications

Enterprises should determine whether systems fall within generative-AI, high-impact-AI or advanced-AI provisions under Korean statutory definitions. Applicable controls may include notice and labelling, risk management, documentation, user protection, human oversight, safety evaluation and evidence retention. Cross-border providers should review territorial scope where Korean users or markets are affected.

Read full enterprise implications

Major policy and regulatory signals

  • Framework Act on the Development of Artificial Intelligence and Establishment of Trust
  • Enforcement Decree of the AI Basic Act
  • AI Basic Act Article 31 transparency duties
  • AI Basic Act Article 34 high-impact AI duties
  • AI Basic Act Article 35 impact assessment
Browse policies and frameworks

Key institutions

  • Ministry of Science and ICT
  • National AI governance bodies created under the AI Basic Act
  • Ministry of the Interior and Safety
  • Korean standards and AI safety institutions

Watchlist

What We're Monitoring

Upcoming milestones, unresolved regulatory questions and implementation developments that may change the governance landscape.

  • Enforcement practice and administrative fines under the AI Basic Act
  • High-impact AI designation and interpretation
  • Transparency and synthetic-content labelling implementation
  • Advanced-AI safety threshold practice
  • Public-sector impact-assessment implementation after 2026-08-28
  • Article 31 transparency practice and workforce capability requirements
  • MOIS public-sector AI ethics self-assessment uptake, training and framework revisions; keep this non-binding framework distinct from statutory impact-assessment obligations
  • Follow-on Japan–Korea AISI multilingual evaluation, benchmarks and multimodal red-teaming outputs after the 8–9 October workshop reported on 2026-10-16

Executive summary

The AI Basic Act entered its main implementation phase on 22 January 2026 and establishes a national framework for AI development and trust. It contains binding transparency duties for specified generative and high-impact AI uses, obligations for high-impact systems, advanced-model safety provisions, institutional mechanisms and enforcement tools. Selected amendments took effect in July 2026, while subordinate legislation further defines implementation. On 28 August 2026, amendments to the AI and Data-Based Administration Act and related subordinate rules added a substantial public-sector governance layer. Enterprises should map whether systems fall within high-impact, generative or advanced-AI provisions and design documentation, human oversight, risk management and transparency into deployment processes.

Governance architecture

South Korea’s architecture combines a binding national AI framework statute with subordinate rules and a separate public-sector AI/data administration layer. Article 31 transparency, Article 34 high-impact AI duties, advanced-AI safety provisions and public-sector impact-assessment mechanisms are distinct legal layers and should not be collapsed into an EU-style single risk taxonomy.

MOIS’s 1 October public-sector ethics framework complements the statutory layer through six values, 68 self-check items and three tailored levels of assessment. Its voluntary self-assessment is not a new statutory duty.

Major policies and frameworks

PolicyIssuerYearStatusSummary
Framework Act on the Development of Artificial Intelligence and Establishment of TrustRepublic of Korea / Ministry of Science and ICT2025–2026Binding national AI framework statuteCore statute for AI development, trust, transparency, high-impact AI, advanced-model safety, institutional governance and enforcement.
Enforcement Decree of the AI Basic ActRepublic of Korea2026Binding Presidential Decree / subordinate legislation — current version in force 2026-08-20Binding Presidential Decree, current version in force from 2026-08-20. Article 25 permits a business to request high-impact AI classification confirmation from the Minister of Science and ICT, submitting the product/service overview, development/training-data overview, materials demonstrating use process/results and other relevant documents. This confirmation procedure is distinct from Article 23 transparency, Article 27 high-impact business duties and Article 28 impact-assessment provisions.
AI Basic Act Article 31 transparency dutiesRepublic of Korea2026Binding statutory transparency obligationsCovers prior notice and disclosure duties for specified generative-AI and high-impact uses, including synthetic-media transparency where applicable.
AI Basic Act Article 34 high-impact AI dutiesRepublic of Korea2026Binding high-impact AI dutiesRequires applicable operators to address risk management, explainability, user protection, human management or oversight and documentation.
AI Basic Act Article 35 impact assessmentRepublic of Korea2026Statutory impact-assessment provision - wording and scope must be applied preciselyProvides an impact-assessment mechanism; do not treat its wording as identical to the mandatory Article 34 duties.
Advanced AI safety provisions and threshold rulesRepublic of Korea2026Binding safety and implementation rules where statutory thresholds applyCurrent decree materials define a covered advanced-AI system using cumulative training compute of at least 10^26 FLOPs together with additional state-of-the-art and serious-risk conditions; this is not a universal definition of all frontier AI.
AI and Data-Based Administration Activation ActRepublic of Korea / Ministry of the Interior and Safety2026Binding public-sector AI and data administration statuteCreates the public-sector governance layer for AI and data administration, including public-body management and impact-assessment mechanisms effective 28 August 2026.
Public-sector AI ethics standardsMinistry of the Interior and Safety2026Final public-sector ethics framework / non-binding self-assessment guidanceMOIS announced final public-sector AI ethics standards on 1 October 2026: six core values, 68 detailed check items and a three-stage tailored self-assessment approach. The framework is practical, non-binding public-sector guidance and voluntary self-checking, not new general legislation or the binding AI Basic Act and Enforcement Decree.

Policy timeline

Milestones are grouped relative to the last review; source date precision is retained.

Recorded milestones

2025-01-21

AI Basic Act promulgated

South Korea promulgated its dedicated national AI framework statute.

2026-01-22

Main AI Basic Act commencement

The AI Basic Act entered its main implementation phase.

2026-07-21

Selected AI Basic Act amendments take effect

Selected amended provisions and current consolidated-law requirements take effect.

2026-08-20

Current Enforcement Decree version reflected

Binding Presidential Decree, current version in force from 2026-08-20. Article 25 permits a business to request high-impact AI classification confirmation from the Minister of Science and ICT, submitting the product/service overview, development/training-data overview, materials demonstrating use process/results and other relevant documents. This confirmation procedure is distinct from Article 23 transparency, Article 27 high-impact business duties and Article 28 impact-assessment provisions.

2026-08-28

Public-sector AI and data administration amendments effective

The AI and Data-Based Administration Act amendments and related subordinate rules add a substantial public-sector governance layer.

2026-10-01

MOIS finalises public-sector AI ethics standards

MOIS announced final public-sector AI ethics standards on 1 October 2026: six core values, 68 detailed check items and a three-stage tailored self-assessment approach. The framework is practical, non-binding public-sector guidance and voluntary self-checking, not new general legislation or the binding AI Basic Act and Enforcement Decree.

2026-10-16

Japan AISI reports first joint technical workshop with Korea AISI

Japan AISI reported the first Japan–Korea joint technical workshop on 16 October 2026; the event itself took place on 8–9 October with 30 participants. Topics included multilingual/multicultural evaluation, sectoral safety methods, JAI-Trust benchmarks and Korean multimodal red teaming. This is technical cooperation and testing capacity, not a new law, binding standard or memorandum.

Enterprise implications

Enterprises should determine whether systems fall within generative-AI, high-impact-AI or advanced-AI provisions under Korean statutory definitions. Applicable controls may include notice and labelling, risk management, documentation, user protection, human oversight, safety evaluation and evidence retention. Cross-border providers should review territorial scope where Korean users or markets are affected.

Classification confirmation is a separate procedure: under Decree Article 25 a business may request confirmation from the Minister of Science and ICT with a product/service overview, training-data overview, evidence of use process/results and other relevant documents. Keep that classification request separate from Article 23 transparency, Article 27 high-impact business duties and Article 28 assessment provisions.

Observatory interpretation

South Korea’s model is a dedicated framework statute with industrial-development and trust objectives, supplemented by detailed subordinate rules and public-sector administration law. It is binding and materially consequential, but its categories and duties should be interpreted from Korean statutory text rather than imported from the EU AI Act.

Official resources

ResourceSourceTypeDateLegal forceWhy it matters

Official current record of South Korea’s national AI framework statute.

National Law Information CenterLaw2025-01-21 / effective 2026-01-22BindingPrimary source for national AI duties, institutions and trust governance.

Binding Presidential Decree, current version in force from 2026-08-20. Article 25 permits a business to request high-impact AI classification confirmation from the Minister of Science and ICT, submitting the product/service overview, development/training-data overview, materials demonstrating use process/results and other relevant documents. This confirmation procedure is distinct from Article 23 transparency, Article 27 high-impact business duties and Article 28 impact-assessment provisions.

National Law Information CenterLaw2026BindingProvides the binding high-impact classification-confirmation procedure; read alongside the separate transparency, business-duty and assessment provisions.

Official transparency provision for specified AI uses.

National Law Information CenterLaw2026BindingSupports precise analysis of notice and disclosure requirements.

Official records for high-impact AI duties and impact-assessment provisions.

National Law Information CenterLaw2026BindingPreserves the distinction between mandatory Article 34 controls and Article 35 assessment wording.

Official public-sector AI and data administration statutory record.

National Law Information CenterPublic-sector rule2026-08-28BindingAdds the public-sector governance and impact-assessment layer.

Specifies notice and labelling methods for relevant high-impact and generative-AI products/services, with stated exceptions.

Republic of Korea / Ministry of Science and ICTLaw2026-08-18BindingRead the binding subordinate provision alongside its parent Act; maintain the distinction between separate procedures and duties.

A business may request ministerial confirmation with product/service, training-data, use-process/results and other relevant materials.

Republic of Korea / Ministry of Science and ICTLaw2026-08-18BindingRead the binding subordinate provision alongside its parent Act; maintain the distinction between separate procedures and duties.

Separately specifies public disclosure of relevant risk management, explanation, user protection and oversight information, with trade-secret exceptions, and five-year evidence retention.

Republic of Korea / Ministry of Science and ICTLaw2026-08-18BindingRead the binding subordinate provision alongside its parent Act; maintain the distinction between separate procedures and duties.

Specifies assessment content, including affected people/groups, fundamental rights, social/economic effects, use, indicators, mitigation and improvement plans. Assessment may be conducted directly or through a third party; do not convert this into a blanket mandatory pre-deployment duty for every AI system.

Republic of Korea / Ministry of Science and ICTLaw2026-08-18BindingRead the binding subordinate provision alongside its parent Act; maintain the distinction between separate procedures and duties.

MOIS announced final public-sector AI ethics standards on 1 October 2026: six core values, 68 detailed check items and a three-stage tailored self-assessment approach. The framework is practical, non-binding public-sector guidance and voluntary self-checking, not new general legislation or the binding AI Basic Act and Enforcement Decree.

Ministry of the Interior and SafetyFramework2026-10-01GuidanceAdds public-sector implementation guidance without changing the force of the separate statutes.

Japan AISI reported the first Japan–Korea joint technical workshop on 16 October 2026; the event itself took place on 8–9 October with 30 participants. Topics included multilingual/multicultural evaluation, sectoral safety methods, JAI-Trust benchmarks and Korean multimodal red teaming. This is technical cooperation and testing capacity, not a new law, binding standard or memorandum.

Japan AI Safety InstituteReport2026-10-16Not applicableOne shared event with separate event and publication dates; no new legal obligation.