What this topic means

Korea AI policy and regulation refers to the laws, government programs, technical standards, and enforcement practices that shape how artificial intelligence is developed and used in South Korea.

The country's central framework is officially titled the Framework Act on the Development of Artificial Intelligence and the Establishment of a Foundation for Trustworthiness. It is commonly called the AI Framework Act or AI Basic Act.

The Act was enacted in 2025 and, together with its Enforcement Decree and key MSIT guidelines, took effect on January 22, 2026. It is designed to promote the AI industry while establishing baseline obligations for transparency, safety, reliability, and user protection.

South Korea has therefore moved beyond voluntary AI guidance toward a comprehensive legal framework — but one still being operationalized through ministerial guidance, technical interpretations, and an active grace period. The framework is live; how far its specific obligations reach is being settled in practice rather than in the statute.

This page is general information about a developing legal framework, not legal advice. Compliance questions require Korean counsel.

Why it matters

South Korea is one of the world's most important technology economies, with major positions in semiconductors, memory, smartphones, telecommunications, robotics, automobiles, batteries, gaming, and consumer electronics.

Its AI law matters because it applies to both AI developers and businesses that deploy AI in products or services. That makes the framework relevant not only to model companies, but also to banks, hospitals, manufacturers, telecom operators, online platforms, software companies, and foreign providers serving Korean users.

For Western readers, Korea's AI policy is important because:

• It is one of the first countries after the European Union to put a comprehensive AI framework into force.
• It combines industrial promotion with enforceable governance obligations — and the balance between the two is visible in how lightly it is being enforced.
• It offers a third path between the EU's detailed risk regime and the voluntary approaches taken in Japan and Singapore.
• Its rules may influence other technology-oriented economies in Asia.
• Its implementation shows how a major manufacturing and technology country regulates AI without attempting to slow it down.

Current legal status

The AI Framework Act is in force

The AI Framework Act and its Enforcement Decree became effective on January 22, 2026, establishing the institutional structure for AI promotion, governance, transparency, safety, and trustworthiness.

It is not simply an AI-safety law. It also directs the government to support research, investment, talent development, infrastructure, commercialization, and the growth of the AI industry — an emphasis written into the title itself.

Implementation is still developing

The Act establishes broad obligations, but practical details depend on the Enforcement Decree, ministerial guidance, official interpretations, and sector-specific implementation.

The Ministry of Science and ICT (MSIT) has implemented a grace period running through January 22, 2027. During this period, fact-finding investigations and administrative fines are deferred, except in serious cases involving significant social harm such as loss of life or fundamental human-rights violations.

The grace period is not a suspension of the law — the obligations exist and apply. What is deferred is the consequence of failing them. Under Article 43 of the AI Framework Act, non-compliance carries a maximum administrative fine capped at KRW 30 million (approximately $21,000 USD) (Source: MSIT / Article 43) for breaches such as failing to provide user notices, omitting generative AI labels, or failing to appoint a required domestic representative. This represents a stark contrast to the European Union's AI Act, which relies on turnover-linked penalties reaching up to €35 million or 7% of a company's total worldwide annual turnover. Korea’s choice of a fixed, modest financial penalty underscores its strategy to prioritize industry promotion and voluntary compliance over aggressive financial deterrence.

CONTRAST: Unlike the EU AI Act—which establishes turnover-based penalties reaching up to €35 million or 7% of global turnover—the Korean Act relies on fixed, low-ceiling administrative fines to encourage compliance without imposing punitive financial stress on industry players.

Extraterritorial reach

The framework reaches foreign companies whose AI products or services affect users or markets in South Korea. Foreign AI businesses without a Korean office must designate a domestic representative if they meet specified thresholds:

• Global annual revenue of KRW 1 trillion or more
• Domestic AI sales of KRW 10 billion or more
• At least 1 million daily active users in South Korea

At August 2026 exchange rates, KRW 1 trillion is roughly USD 680 million and KRW 10 billion roughly USD 6.9 million.

Main regulatory concepts

AI Developers vs. AI Deployers

The Act distinguishes between entities that develop AI models (AI Developers) and businesses that integrate or provide services using AI (AI Deployers).

This distinction carries most of the framework's practical reach. A company does not need to build its own foundation model to fall under the Act: a bank, hospital, manufacturer, or software company has deployer obligations if it uses AI in a customer-facing product or operational service.

Generative AI

Generative AI refers to systems that generate text, images, audio, video, software, or other content.

Operators of generative AI systems must provide appropriate notice to users and, in relevant cases, label AI-generated outputs. Deepfake and synthetic-media outputs receive additional attention, on the reasoning that users may otherwise mistake them for authentic human-created content.

High-impact AI

High-impact AI refers to systems that may significantly affect human life, physical safety, or fundamental rights.

The concept applies most directly in healthcare, employment, education, public services, finance, energy, and transportation — settings where an AI decision materially affects a person or a critical service.

Providers of high-impact AI face stronger requirements than providers of lower-risk systems, including risk management, explanation of outcomes, training-data information, safety measures, and user-protection procedures.

High-performance AI

The Enforcement Decree identifies a high-performance category tied to advanced model development and large-scale computing. Reports and guidance describe a cumulative training-compute threshold of 10^26 floating-point operations (FLOPs) for enhanced safety obligations — a figure that echoes thresholds used in US and EU frontier-model policy.

The precise technical classification and its practical application continue to be settled through MSIT guidance.

Main obligations

Advance notice

Businesses must notify users when a product or service uses generative AI or high-impact AI in circumstances covered by the framework. The notice must be understandable and provided before or at the relevant point of use; a general privacy-policy mention is not treated as sufficient.

Output labeling

Certain content produced by generative AI must be identified as AI-generated, with deepfake outputs receiving particular attention because they can mislead users about whether a person, event, or statement is authentic.

The technical method depends on the content type and on MSIT guidance. The open question is durability — whether labels survive when content is shared, cropped, or re-encoded.

Risk management

Providers of high-impact AI must establish and operate risk-management procedures covering the full life cycle of the system: development, testing, deployment, monitoring, incident response, and modification. In practice, this means identifying foreseeable harms, naming responsible personnel, and defining escalation paths and controls.

Explainability

High-impact AI providers may need to provide a meaningful explanation of an outcome, including the criteria or principles that influenced it.

What counts as meaningful varies by system and context, and the gap matters: a technical description of a model is not the same thing as an explanation a user, affected person, regulator, or business customer can act on.

Training-data information

The framework requires information about training data or the principles used to develop a system, particularly for high-impact AI — data sources, governance, quality checks, known limitations, and changes made during development.

User protection

AI operators must take steps to protect users from foreseeable harm. Depending on the system, that can involve human review, complaint procedures, correction mechanisms, access controls, monitoring, and the ability to suspend or modify an unsafe service.

How Korea's approach compares

South Korea sits between the European Union's prescriptive risk regime and the voluntary frameworks favored in Japan and Singapore: comprehensive statute and real obligations, but written with industrial promotion in the title and enforced with a year-long grace period.

China occupies different ground again, with state-led algorithm governance and content controls that Korea's framework does not attempt to replicate.

The five-market comparison — Japan, China, South Korea, Singapore, and Taiwan — is set out in Asia AI Regulation Tracker.

Policy and industry priorities

AI industry promotion

The framework is intended to make South Korea a stronger AI power, not merely to constrain AI companies. Government policy supports research, commercial deployment, infrastructure, talent, and partnerships between technology companies and established industries.

AI safety and trust

The government is building institutions and processes for AI safety, evaluation, reliability, and public trust — work that grows more consequential as models gain access to business systems and physical infrastructure.

Public and high-impact applications

Healthcare, finance, education, employment, public administration, transportation, and energy receive particular attention because AI decisions in these sectors directly affect rights, safety, or access to essential services.

Global competitiveness

South Korea is trying to avoid a regulatory system that prevents domestic companies from competing internationally — enough trust and accountability to protect users, without foreclosing model development and industrial adoption. Whether that balance holds is the substance of the next two years.

What to watch next

• Enforcement guidance: additional MSIT notices, technical guidance, official interpretations, and sector-specific explanations.
• The grace period: whether the window closing in January 2027 is extended, narrowed, or replaced by active enforcement.
• High-impact classifications: how regulators classify AI used in healthcare, finance, employment, education, public services, energy, and manufacturing.
• Generative-AI labels: how companies implement notices and labels for AI-generated text, images, audio, video, and deepfakes.
• Frontier-model obligations: how the high-performance category (10^26 FLOPs) is interpreted and audited.
• Foreign-company compliance: how international model providers, cloud companies, and software vendors respond to domestic-representative requirements.
• Sector-specific regulation: the Act is a general framework, and additional obligations arise from existing laws on privacy, medical devices, financial services, and telecommunications.

Simple policy map

A simplified view of Korea's AI policy approach looks like this:

• AI promotion supports research, investment, infrastructure, talent, and commercialization.
• General transparency requires notice when covered products or services use AI.
• Generative-AI rules address user notification and labeling of certain generated content.
• High-impact AI rules add risk management, explanations, training-data information, and user protections.
• High-performance AI rules introduce additional safety expectations for advanced systems above 10^26 FLOPs.
• Government oversight is led primarily by MSIT, alongside sectoral authorities.
• Transition and enforcement are phased, with a general grace period through January 2027 for fines and investigations except in serious-harm cases.

Related AsiaAI coverage

• Asia AI Regulation Tracker
• Japan AI Policy and Regulation
• Korea AI Industry Map
• Korea Semiconductor Ecosystem
• China AI governance and enforcement
• How Korean banks and manufacturers are deploying enterprise AI
• Korea's AI safety, model-evaluation, and frontier-compute programs
• The effect of Korean AI rules on foreign model providers

FAQ

What is South Korea's AI Framework Act?

South Korea's comprehensive legal framework for promoting artificial intelligence and establishing trustworthiness requirements. It covers AI-industry development, transparency, safety, high-impact systems, generative AI, and user protection.

When did the Korean AI Act take effect?

The Act and its Enforcement Decree took effect on January 22, 2026. Implementation details continue to develop through guidance and administrative practice.

Is South Korea's AI law similar to the EU AI Act?

Both are comprehensive AI laws with additional obligations for higher-risk systems, but they are not the same instrument. Korea's framework places a stronger explicit emphasis on AI-industry promotion and is generally less prescriptive.

What is high-impact AI in Korea?

AI that may significantly affect human life, physical safety, or fundamental rights — including systems used in healthcare, finance, employment, education, public services, and energy.

What does the law require for generative AI?

Covered operators must notify users that generative AI is in use, and certain AI-generated or deepfake content must be labeled. The method depends on the system and applicable implementation guidance.

Does the law apply to foreign companies?

It applies to foreign companies whose AI products or services affect Korean users or markets. Foreign operators meeting the global revenue (KRW 1 trillion+), domestic AI revenue (KRW 10 billion+), or daily user (1M+) thresholds must appoint a domestic representative.

Is the Korean AI Act currently being enforced?

The Act is in force, but administrative fines and non-urgent investigations are subject to a grace period through January 22, 2027, except in serious cases involving significant social harm.

What should Western readers watch next?

MSIT guidance, high-impact classifications, generative-AI labeling practices, frontier-model thresholds, sector-specific rules, and whether enforcement actually begins in 2027.

Last updated: August 2026. This page will be updated as the AI Framework Act, Enforcement Decree, guidance, and enforcement practices develop.