Status at a glance
The AI Basic Act and its Enforcement Decree took effect on January 22, 2026 (US International Trade Administration). Reuters reported that the Ministry of Science and ICT described the package as the world's first complete set of AI laws to take effect in full, ahead of the EU AI Act, which applies in phases (Reuters, via BusinessDay). Companies have a grace period of at least one year before administrative fines are imposed, and the ministry said it was considering an extension if industry conditions warranted it (same source). As of November 9, 2026, we found no report of a fine under the Act.
The Act has been amended twice since enactment. Act No. 21311, promulgated January 20, 2026 and effective January 22, 2026, renamed the National AI Committee as the National AI Strategy Committee, expanded support for training data and AI research institutes, and requires that high-impact AI impact assessments reflect the characteristics of AI-vulnerable groups. Act No. 22076, promulgated October 20, 2026, creates an integrated system for AI training data and takes effect April 21, 2027 (law.go.kr, Act No. 21311; law.go.kr, Act No. 22076). Neither amendment changed the core transparency or safety duties described below.
What is the Korea AI Basic Act?
The AI Basic Act, formally the Framework Act on the Development of Artificial Intelligence and the Establishment of Trust, is South Korea's general AI statute. The National Assembly passed it in December 2024 (IAPP), and it was promulgated as Act No. 20676 on January 21, 2025, with a one-year lead time (law.go.kr). The government's stated purpose is to support the healthy development of AI and build a foundation of trust, protecting citizens' rights and dignity while strengthening national competitiveness (same source).
It is two laws in one. Much of the Act is industrial policy: a national AI master plan every three years, a presidential committee, an AI Policy Center and an AI Safety Research Institute, and support for research, data centers, training data, startups and talent (Cooley). The regulatory part is narrower and sits in three places: transparency for generative AI, obligations for high-impact AI, and safety duties for the most powerful models. The Enforcement Decree, issued by MSIT, supplies the thresholds and procedures.
Critics have read that balance as tilted toward promotion. Writing for the IAPP before the decree was final, Kyoungsic Min argued that the grace period, the modest fine ceiling and high thresholds left the Act with little deterrent effect, and that Korea's privacy regulator remains the more powerful enforcer (IAPP). Startup groups, by contrast, worried that vague terms would push them to cautious, less innovative designs (Reuters).
Does the South Korea AI Act apply to foreign companies?
Yes. The Act applies to two kinds of AI business operator: those that develop and provide AI, and those that provide products or services using AI developed by others. The categories do not map neatly onto the EU's provider and deployer, and one company can be both. The Act applies to activity outside Korea that affects the Korean market or users, with an exception for AI used for national defense or security (Cooley).
Foreign AI businesses with no address or office in Korea must appoint a domestic representative if they meet any one of three thresholds in the Enforcement Decree:
- total revenue of at least KRW 1 trillion in the previous year;
- revenue from AI services of at least KRW 10 billion in the previous year; or
- an average of at least one million daily users in Korea over the three months before the end of the previous year.
The representative answers government requests and handles safety reporting (US International Trade Administration; Cooley). Under the decree, a company that uses an AI system whose developer has already met the relevant obligations generally does not need to take separate measures (IAPP).
What does the AI Basic Act require?
Transparency and labeling for generative AI
Operators that offer high-impact or generative AI must notify users in advance that the product or service uses AI. Operators of generative AI must label output as AI-generated, and where AI-generated sound, images or video are difficult to distinguish from real ones, they must give clear notice that the content is AI output (Cooley). MSIT guidance distinguishes output that stays inside a service, such as a chatbot or game, where a notice in the interface, a logo or a pop-up can suffice, from content that can be downloaded or shared, which is expected to carry clearer labeling or identification (US International Trade Administration).
High-impact AI
High-impact AI is AI that can significantly affect life, physical safety or fundamental rights when used in listed areas: energy supply; drinking water production; healthcare systems; medical devices and digital medical equipment; nuclear materials; biometric analysis for criminal investigation; evaluations with major effects on rights and obligations, such as hiring and credit assessment; transport; public-sector decisions on services and fees; student assessment in early childhood, primary and lower secondary education; and other areas with serious effects on life, safety or rights (summary of Article 2 published by Vietnam's Ministry of Justice portal). The decree did not add areas to the statutory list (IAPP).
Operators must assess before deployment whether their system is high-impact and may ask MSIT to confirm. If it is, they must provide a meaningful explanation of outcomes, the main criteria used and a summary of training data; create a user protection plan; build in human oversight and intervention; and document their trust and safety measures. They must also make efforts to assess the system's impact on fundamental rights before use (Cooley).
High-performance AI
Systems trained with cumulative compute of at least 1026 floating-point operations are designated high-performance AI. Their operators may be required to run a life-cycle risk management plan and user protection measures and to report results to MSIT (Cooley). The US Trade Administration describes the criteria as combining the compute threshold with use of state-of-the-art technology and potential broad impact on fundamental rights (trade.gov).
When did South Korea's AI law take effect, and what is the grace period?
The Act took effect on January 22, 2026, one year after promulgation, and the Enforcement Decree took effect the same day (law.go.kr; trade.gov). Unlike the EU AI Act, there was no staggered schedule: transparency, high-impact and high-performance duties all applied from that date.
The softening comes through enforcement. MSIT committed to a grace period of at least one year before it imposes administrative fines, and planned a guidance platform and support center for companies during that period (Reuters). The earliest point at which fines could begin is therefore January 2027, but the ministry has not fixed an end date and has said it may extend the period. The grace period covers fines; it does not suspend the duties themselves, and MSIT's power to order corrections or suspend a service that threatens safety remains (Cooley).
What are the penalties under the AI Basic Act?
The Act's fines are administrative and capped at KRW 30 million, about US$21,000 according to Cooley. They apply to three failures: not notifying users that AI is in use, not appointing a required domestic representative, and violating corrective orders or refusing inspections (Cooley). Reuters gave the example that failing to label generative AI could bring a fine of up to KRW 30 million (Reuters).
MSIT is the regulator. It can investigate, order corrective action and order a service suspended where it poses a threat to safety. Data protection questions raised by AI systems, such as training on personal data, fall under Korea's Personal Information Protection Act and its own regulator, which the IAPP piece describes as the stronger enforcer.
How it compares with the EU AI Act
Both laws are risk-based and both reach foreign companies, but they differ in reach, timing and bite. The table compares the main points as of November 9, 2026; the EU column draws on Regulation (EU) 2024/1689 and our EU AI Act overview.
| Feature | Korea AI Basic Act | EU AI Act |
|---|---|---|
| In force | All duties from Jan. 22, 2026 | Phased: bans 2025, transparency Aug. 2026, high-risk Dec. 2027 and Aug. 2028 |
| Banned practices | No list of prohibited AI practices | Article 5 prohibitions |
| Risk category | High-impact AI in listed areas | High-risk AI under Annexes I and III |
| Large-model threshold | 1026 FLOPs (high-performance AI) | 1025 FLOPs presumption for systemic risk |
| AI content labeling | Label generative output; clear notice for realistic synthetic media | Article 50 marking and deepfake labels |
| Local representative | Foreign firms above revenue or user thresholds | Required for non-EU providers of high-risk systems and general-purpose models |
| Maximum fine | KRW 30 million | EUR 35 million or 7% of worldwide turnover |
| Fines active | After a grace period of at least one year | Yes, for obligations already applicable |
The IAPP piece noted that Korea's compute threshold is ten times the European one, so very few models are likely to qualify (IAPP). For a company already building to the EU's Article 50 marking and labeling rules, Korea's labeling duty will look familiar; the Korea-specific work is Korean-language notices, the high-impact self-assessment and, for large foreign firms, the local representative. Other national AI laws are collected on our international AI laws page.
South Korea's AI Basic Act has applied in full since January 22, 2026: advance notice and labels for generative AI, oversight and explanation duties for high-impact AI, and local representatives for large foreign firms. Fines are capped at KRW 30 million and, as of November 9, 2026, remain on hold during a grace period of at least one year, so the near-term risk is corrective orders rather than penalties.
What changed
- 2026-11-14: Page published, with status verified as of November 9, 2026.
- 2026-10-20: Amendment Act No. 22076 promulgated, creating an integrated AI training data system; effective April 21, 2027.
- 2026-01-22: AI Basic Act and Enforcement Decree took effect; grace period of at least one year for fines began.
- 2026-01-20: Amendment Act No. 21311 promulgated, renaming the National AI Committee and expanding support provisions; effective January 22, 2026.
- 2025-01-21: AI Basic Act promulgated as Act No. 20676.
Frequently asked questions
What is the Korea AI Basic Act?
It is South Korea's general AI law, the Framework Act on the Development of Artificial Intelligence and the Establishment of Trust. It combines industrial support for AI with duties on AI businesses: advance notice and labeling for generative AI, safety and oversight measures for high-impact AI, and risk management for the largest models.
When did South Korea's AI law take effect?
The AI Basic Act took effect on January 22, 2026, together with its Enforcement Decree, one year after it was promulgated on January 21, 2025. Administrative fines are subject to a grace period of at least one year, so the earliest they could begin is January 2027.
Does the South Korea AI Act apply to foreign companies?
Yes. It applies to AI activity outside Korea that affects the Korean market or users. Foreign firms without a Korean office must appoint a local representative if they had KRW 1 trillion in total revenue, KRW 10 billion in AI revenue, or one million average daily Korean users.
What are the penalties under the AI Basic Act?
Administrative fines of up to KRW 30 million for failing to notify users of AI use, failing to appoint a required domestic representative, or violating corrective orders. MSIT can also order corrections or suspend unsafe services. Fines are not being imposed during a grace period of at least one year.
Sources
- Korea Law Information Center: AI Basic Act, Act No. 20676 (enactment)
- Korea Law Information Center: Amendment Act No. 21311 (Jan. 20, 2026)
- Korea Law Information Center: Amendment Act No. 22076 (Oct. 20, 2026)
- US International Trade Administration: South Korea AI Basic Act
- Cooley: South Korea's AI Basic Act overview and key takeaways (Jan. 27, 2026)
- Reuters via BusinessDay: South Korea launches landmark laws to regulate AI (Jan. 22, 2026)
- IAPP: South Korea's AI Act designed to be all roar, no bite (Sept. 10, 2025)
- Vietnam Ministry of Justice portal: The Framework Act on Artificial Intelligence of the Republic of Korea
- EUR-Lex: Regulation (EU) 2024/1689 (AI Act)
This page is journalism and commentary, not legal advice. AI law changes quickly; check the primary sources above and consult qualified counsel for your situation.