.svg){kind=logo}
KRW 30 million, about US$20,000 to US$21,000. That is the most any single violation can cost under the South Korea AI Framework Act, and it is the least interesting number in it. The part that matters to a company outside Korea is who counts as a regulated operator: any business whose AI affects Korean users or the Korean market.
The South Korea AI Framework Act, widely called the AI Basic Act, is a comprehensive AI statute in force since 22 January 2026. It applies to AI that affects Korean users or markets, sets duties for high-impact AI, requires labelling of generative AI output and caps fines at KRW 30 million. The ministry has paused fines for at least a year.
The duties themselves are already in force. Only the fines are paused, and that pause is a ministry policy announcement, not a statutory moratorium. Large foreign operators must also name a domestic representative in Korea.
Quick take
If your AI product has Korean users, check three things first: whether you cross a domestic representative threshold, whether any system is high-impact, and whether your generative AI output is labelled. Everything else in the Act follows from those three.
By the end you will know whether you are in scope, which provision can reach you first, what the grace period does and does not cover, and what to put in place now.
Here's what I'll cover:
- The law, its two names and where it stands in October 2026
- Who it reaches, and the domestic representative thresholds
- High-impact AI and generative AI labelling duties
- Penalties, the grace period and how the Act compares with the EU AI Act
- A compliance checklist, with ISO 42001 as a baseline
This article covers South Korea's statute only. For the multi-country view, see our AI regulations by country map, and for the EU's law see the EU AI Act compliance guide.
What the South Korea AI Framework Act Is and Where It Stands in October 2026
The Korea Legislation Research Institute's English text gives the formal title as the Framework Act on the Development of Artificial Intelligence and the Creation of a Foundation for Trust, Act No. 20676, promulgated on 21 January 2025. One statute does two jobs: it promotes the AI industry and it regulates how AI is built and used.
Three layers sit inside it:
- Promotion: a national AI strategy committee under the President, support for data and infrastructure, and measures aimed at SMEs and startups.
- Trust and safety duties: user notice, generative AI labelling, high-impact AI measures and safety duties for very large models.
- Enforcement: ministry investigations, corrective orders and administrative fines.
Most readers outside Korea only need the second and third layers.
The South Korea AI Basic Act and the South Korea AI Framework Act: One Law, Two Names
They are the same statute, and the South Korea AI Basic Act label is simply the more common one abroad. The Korean title uses the word for "basic," and Georgetown's CSET translation notes that a more literal English rendering is "Basic Act." Law firms, the US International Trade Administration and the Korean government's own English pages mostly say AI Basic Act. FPF and several others say AI Framework Act.
Machine translations produce other variants, such as "AI Development and Trust Act." If a page uses one of those names and cites Act No. 20676, it is describing this law. I use "AI Framework Act" for the formal name and "AI Basic Act" where a source does.
Quick take
Searching either name leads to the same obligations, the same number and the same regulator, the Ministry of Science and ICT (MSIT).
South Korea AI Law Timeline: From Passage to the End of the Fine Grace Period
Three clocks run at once under the South Korea AI law: the duties, the fines and the guidelines. Mixing them up is how companies decide they have a year to do nothing.
| Date | Event | Status |
|---|---|---|
| December 2024 | National Assembly passes the Act | Done |
| 21 January 2025 | Promulgated as Act No. 20676 | Done |
| 12 November to 22 December 2025 | MSIT public notice on the draft Enforcement Decree, which also announced a grace period of at least one year before administrative fines | Done |
| 20 January 2026 | First amendment promulgated as Act No. 21311 | In the consolidated text |
| 21 January 2026 | Enforcement Decree promulgated (Presidential Decree No. 36053, per DLA Piper's tracker) | Done |
| 22 January 2026 | Act and decree in force | Duties apply now |
| 20 July 2026 | Decree amended (Presidential Decree No. 36506, as reported by one tracker) | Contents unconfirmed |
| 13 October 2026 | Cabinet approves a further amendment bill on AI training data, as reported by Asiae and MLex | Not yet law |
| At least one year from January 2026 | MSIT's grace period before administrative fines | Policy, no official end date |
The first amendment touched Articles 2, 3(5), 6 to 8, 10, 14 to 18, 22-2 and 22-3, plus part of Article 35(1). The October bill concerns training data created through government projects, which is separate from the duties in this article.
Watch out
The grace period has no official end date. Many pages print "22 January 2027," but MSIT's own wording is "at least one year," and Reuters reported in January 2026 that the ministry was considering extending it. No extension had been announced as of 21 October 2026.
The South Korea AI Enforcement Decree and What It Settled
The South Korea AI enforcement decree is where the numbers live, because the Act left them to a Presidential Decree. The decree fills in three things that matter to foreign operators: the domestic representative thresholds, the training-compute level at which very large models carry extra safety duties (10^26 floating-point operations, per MSIT's public notice and Shin & Kim's summary of the final text), and the procedure for confirming high-impact status.
MSIT also issued guidelines, including Transparency Guidelines and guidance on how to decide whether a system is high-impact. Those are ministry guidance, not law, and MSIT has said it will revise the labelling guidance using feedback from the grace period (Kim & Chang, 27 January 2026).
Does the South Korea AI Framework Act Apply to Your Company?
The useful question is not "are we a Korean company." It is "does our AI touch a Korean user or the Korean market." Article 4(1) says the Act applies to conduct outside Korea "if the conduct impacts the domestic market or users." Article 4(2) carves out AI used solely for national defence or security, as set by decree.
The Act regulates "AI business operators" in two roles under Article 2(7): those who develop AI, and those who provide a product or service using AI built by someone else. A SaaS company that calls a third-party model is the second kind.
| Situation | In scope? | What applies first |
|---|---|---|
| Korean companies use your SaaS and its AI features | Likely yes, because Korean users are affected | User notice, labelling, high-impact check |
| Consumers in Korea use your app | Likely yes | Same, plus representative thresholds |
| Korean-language website, no Korean users of the AI | Depends on actual effect on Korean users or market | Document the reasoning |
| Your model is used by a Korean company you never contracted with | Depends; the test is impact on Korean users or the market | Confirm what reaches Korean users |
| AI used only inside your own company, no Korean users | Not on the text of Article 4(1) | Record the determination |
Extraterritorial Reach of the South Korea AI Framework Act Under Article 4
FPF's analysis calls the reach "broad," and the text supports that. What is not yet known is how MSIT will apply it to a company with no Korean footprint. No enforcement action had been reported as of 21 October 2026, so any prediction about first targets is speculation.
What you can do without guessing: decide whether your product affects Korean users, write the reasoning down and date it. That memo is the first document anyone will ask for.
South Korea AI Act Domestic Representative: Three Thresholds and What the Role Does
Article 36 requires a foreign operator with no domicile or place of business in Korea to designate a domestic representative in writing and report it to the minister, if it meets any one of the decree's thresholds.
| Threshold | Test | Rough US equivalent |
|---|---|---|
| Total revenue | At least KRW 1 trillion in the prior year | About US$662 million |
| AI-service revenue | At least KRW 10 billion in the prior year | About US$6.6 million |
| Korean users | At least one million average daily users over the three months before the end of the prior year | Not applicable |
The dollar figures are the US International Trade Administration's conversions, so treat them as approximate. The requirement can also apply to an operator that has been fined for ignoring an MSIT suspension or corrective order (trade.gov).
The South Korea AI Act domestic representative has a practical job under Article 36(1). It receives submissions about very large models, handles confirmation requests about high-impact status and supports high-impact measures. It must be domiciled or have a place of business in Korea, and under Article 36(3) the operator is treated as having committed any violation the representative commits in those matters.
Three consequences for a global company:
- The user test catches small companies. Most startups sit far below the revenue tests, but a popular consumer product can still pass one million daily Korean users.
- A Korean office takes you out of this provision. The duty is for operators without a domicile or place of business in Korea.
- Not designating is one of only three fine triggers. More on that below.
Try this yourself
Pull your last 90 days of daily active users with a Korean locale or IP and average them against one million. Then take your AI-product revenue and compare it with KRW 10 billion at today's exchange rate. Two numbers, about ten minutes, and you know where you stand.
ISO 42001
One AI management system for more than one jurisdiction
Korea is one of several AI laws your customers may ask about. ComplyJet's ISO 42001 workspace helps you keep AI policies, risk and impact assessments and evidence in one place, so each new jurisdiction starts from records you already have. It does not replace legal advice or act as your Korean representative.
South Korea High-Impact AI: Ten Areas, a Two-Part Test and a Prior Review
South Korea high impact AI, written "high-impact AI" in the statute, is the category that carries real duties. Article 2(4) defines it, in paraphrase, as an AI system likely to have a significant impact on, or pose a risk to, human life, physical safety and fundamental rights, used in one of the listed areas.
That is a two-part test: a listed area, plus significant impact or risk. Sitting in a listed area does not make a system high-impact by itself. MSIT's own notice describes the assessment as based on detailed criteria, including application area, risk to basic rights, severity and frequency.
| Listed area (Article 2(4)) | Plain example |
|---|---|
| Energy supply | Grid and supply management |
| Drinking water production | Treatment process control |
| Health and medical services | Systems used to provide and use care |
| Medical devices, including digital medical devices | Diagnostic software |
| Nuclear materials and facilities | Safety management |
| Biometric analysis for criminal investigation or arrest | Face or fingerprint matching by investigators |
| Judgments that significantly affect individuals' rights and obligations | Hiring and loan screening, the statute's own examples |
| Transport and traffic systems | Operation and management of traffic systems |
| Decisions by the State or public bodies that affect citizens | Qualification checks for public services |
| Evaluation of students in early childhood, primary and secondary education | Automated grading and assessment |
| Other areas set by Presidential Decree | Open category |
For a typical software company, South Korea high impact AI most often means hiring tools, credit scoring and anything feeding a public-sector decision.
Before launch, Article 33 requires an operator to review whether its AI is high-impact. You may ask the minister to confirm, and MSIT's English notice puts the answer window at 30 days, extendable once by 30 more.
What the South Korea AI Framework Act Requires of High-Impact AI Operators
Article 34(1) lists the measures. I describe them rather than quote them:
| Duty | What the Act says | What to produce as evidence |
|---|---|---|
| Risk management | Measures to manage risk | A dated risk assessment per system |
| Explanation | Explain results, main criteria and a training-data overview, to the extent technically feasible | A plain-language explanation note |
| User protection | Measures to protect users | A documented user-protection plan |
| Human oversight | Human management and oversight | A named overseer with authority to intervene |
| Documentation | Records that show the safety measures taken | A retained evidence file |
Verdict: these are familiar governance artefacts, and each one is a record rather than a product feature.
Two related provisions matter. Article 31(1) requires advance notice to users that a product or service runs on high-impact or generative AI, and missing that notice is one of the three fine triggers. Article 35 asks operators to "endeavor" to assess the impact on fundamental rights in advance. That is a soft duty today, though public bodies are told to prefer products that have undergone one.
Very large models get a separate rule. Article 32 adds lifecycle risk management and a report to the minister for AI systems whose cumulative training compute meets the decree threshold, 10^26 floating-point operations. That is frontier-model territory and not where a typical SaaS company sits.
Watch out
Failing an Article 34 measure is not itself a fine. It falls within the ministry's investigation power under Article 40(1), and it reaches a fine only if a corrective order under Article 40(3) is ignored.
South Korea AI Act Labelling: What Generative AI Providers Must Disclose
South Korea AI Act labelling starts with a short statute. Article 31(2) says output from generative AI must be indicated as generated by generative AI. Article 31(3) adds that synthetic sound, images or video that are hard to tell from the real must be notified or indicated in a way users can readily recognise, with room for artistic and creative works. Article 31(4) leaves the details and exceptions to the decree.
The mechanics sit in the Transparency Guidelines, as reported by law firms in January and February 2026. Treat them as guidance that MSIT plans to revise.
| Output type | What the guidance expects |
|---|---|
| Output that stays inside your service | Flexible labelling: pre-use guidance or an on-screen symbol or logo can be enough for interactive services |
| Output that users export or download | A human-recognisable mark such as a visible or audible watermark, or machine-readable metadata plus a text or audio notice that the content is AI-generated |
| Deepfakes | Clearly recognisable labelling, with flexibility for creative works |
| Obvious AI use or internal-only business use | Exempt under the draft guidelines |
Sources: Kim & Chang on the in-service and exported split, Shin & Kim on the sanctions.
Myth debunking
"A missing label is fined directly." NO. Article 43(1) does not list labelling. A labelling failure draws a corrective order, and a fine follows only if the order is ignored.
For a global reader, this belongs to the same family as the EU AI Act's Article 50 and China's labelling measures, so one labelling method can often serve all three.
South Korea AI Act Penalties: Three Fine Triggers and a Corrective Order Ladder
The maximum administrative fine is KRW 30 million, about US$20,000 to US$21,000, per violation, and it applies to three things only under Article 43(1): failing to give the advance notice under Article 31(1), failing to designate a domestic representative under Article 36(1), and failing to comply with a cease or correct order under Article 40(3). The fine is a fixed cap, not a share of turnover.
South Korea AI Act penalties for everything else run through a ladder:
| Violation | First consequence | What turns it into a fine |
|---|---|---|
| No advance notice (Article 31(1)) | Direct fine | Already a trigger |
| No domestic representative (Article 36(1)) | Direct fine | Already a trigger |
| No generative AI label (Article 31(2) or (3)) | Investigation, then corrective order | Ignoring the order |
| High-impact measures not met (Article 34(1)) | Investigation, then corrective order | Ignoring the order |
| Very large model safety report not made (Article 32) | Investigation, then corrective order | Ignoring the order |
| Ignoring a corrective order (Article 40(3)) | Fine | Already a trigger |
Under Article 40, the minister can request data, order investigations, enter premises to inspect records and order an operator to stop or correct a violation.
Verdict: the cost of getting this wrong is mostly the corrective-order process and the work behind it, not the fine itself.
My view is that KRW 30 million is small next to the EU's top tier of up to 7% of worldwide turnover, but the amount is not the point for a company selling to Korean enterprises. A corrective order is a regulator's letter that a customer's security team can ask about.
South Korea AI Act Grace Period: What Is Paused, What Is Not and What Comes Next
MSIT's English notice says: "A grace period of at least one year will be implemented before imposing administrative fines." The South Korea AI Act grace period is wider than the fines alone.
By January 2026, law firms and the US International Trade Administration described the guidance period as also covering fact-finding investigations and corrective orders, with exceptions only for very serious social harm such as loss of life or human rights violations. That is a ministry announcement. It is not a provision of the Act.
| Item | Status |
|---|---|
| The duties (notice, labelling, high-impact measures) | In force and binding |
| The representative requirement | In force and binding |
| Administrative fines | Paused by policy |
| Investigations and corrective orders | Generally deferred, with serious-harm exceptions |
| End date of the grace period | Not officially set |
| Whether it is extended | Not announced |
| Contents of the 20 July 2026 decree amendment | Unconfirmed |
The practical use of a pause is to produce records, not to wait. The Korea AI and Software Industry Association runs an AI Basic Act Help Desk, which Kim & Chang reports answers general questions within 72 hours and complex legal ones within 14 days.
South Korea AI Act vs EU AI Act: Where They Differ for a Global Company
South Korea AI Act vs EU AI Act is not a copy-and-paste comparison: they rhyme, but they are not the same law. Companies that treat one as a copy of the other tend to miss the representative requirement and the notice duty.
| Dimension | South Korea | EU | So what |
|---|---|---|---|
| Approach | Framework and promotion statute with a thin duty layer | Risk-tier regulation with banned practices | Different shapes, same family of controls |
| Prohibited practices | None, per FPF's analysis of the Act | A list of banned uses | No Korean ban list to screen against |
| Risk category | High-impact AI: listed area plus significant impact | Annex III and Annex I high-risk systems | Classify twice; the tests differ |
| Generative AI transparency | Article 31 notice and labelling | Article 50, applying from 2 August 2026 | One labelling method can serve both |
| Fines | KRW 30 million cap, three triggers | Up to EUR 35 million or 7% of worldwide turnover for the top tier | Fines differ by orders of magnitude |
| Foreign operators | Domestic representative above thresholds | Authorised representative for providers outside the EU | Check both |
| Timing | Duties in force; fines paused by MSIT policy | Annex III high-risk from 2 December 2027 and Annex I from 2 August 2028, per the Commission's timeline | Different clocks |
| Regulator | MSIT, with PIPC and others in their domains | AI Office and national authorities | Different counterparties |
The EU dates above come from the Commission's AI Act Service Desk timeline, which reflects the Digital Omnibus amendments (AI Act Service Desk). Our EU AI Act compliance guide article covers the EU side in detail.
The better question is not which law is stricter. It is which duties one process can serve at once. Labelling, human oversight, risk management, documentation and an inventory of AI systems all overlap, and that overlap is where the effort should go.
Who Regulates the South Korea AI Framework Act, and How PIPA Fits In
The Minister of Science and ICT is the authority the Act names for submissions, confirmations, inspections and corrective orders. MSIT also issued the decree and the guidelines.
There is no single supervisory authority across all AI in Korea, so the South Korea AI law sits beside other regulators. DLA Piper's tracker says the minister has overall responsibility for policy and enforcement, while the Personal Information Protection Commission (PIPC), the Korea Media and Communications Commission and the Korea Fair Trade Commission handle issues in their own domains. FPF describes the PIPC as a de facto regulator for AI that uses personal data and flags the coordination gap.
The Act does not replace the Personal Information Protection Act or sector rules. Anything that uses Korean personal data is also a data-protection question, and our cross-border data transfer requirements guide covers transfer mechanics. This article does not analyse PIPA further.
Note
During the pause, the practical contact points are MSIT's briefings and the KOSA-run AI Basic Act Help Desk.
South Korea AI Act Compliance Checklist for Global Companies
South Korea AI Act compliance, with thin penalties and open questions, is mostly a records job. Here is the checklist I would run, with owners as my suggestions rather than anything the Act prescribes.
- Inventory every AI feature and mark which touch Korean users. Start from the owner-first build in our AI governance framework guide, and check that the basics exist before you start.
- Test the three representative thresholds against real numbers. Owner: finance for revenue, product for users.
- Run and date a high-impact determination per product. Decide whether to ask MSIT to confirm. Our AI risk assessment guide covers how to score a system.
- Draft the advance-notice text and decide where it is served. This is a fine trigger, so keep proof it was shown.
- Choose a labelling method for in-service and exported output. Check it against the current Transparency Guidelines.
- Name the human overseer and write the explanation and user-protection plans for any high-impact system.
- Keep the documentation set. Your AI governance policy should say who keeps them and for how long.
- Name who answers a ministry corrective order, and how fast. Ignoring one is a fine trigger.
- Map third-party models and suppliers. See AI vendor risk management.
- Review the position quarterly against guideline revisions. If you already have a committee, it is the natural reviewer, and our AI governance committee guide covers how to set one up.
Decide Where You Stand: A Decision Path for Foreign Companies
Choose the first line that fits:
- No Korean users and no effect on the Korean market: record the reasoning and re-check when you launch there.
- Korean users, and a system in a listed high-impact area: operate the Article 34 measures and consider asking MSIT to confirm.
- Korean users and generative AI output: serve the Article 31 notice and label the output.
- Above a representative threshold with no Korean office: designate a domestic representative and report it.
- None of the above: write a dated determination memo for each product, and review it each quarter.
ISO 42001 and the South Korea AI Framework Act: A Baseline, Not a Safe Harbour
ISO 42001 does not make a company compliant with Korean law. Certification does not confer legal compliance with Korea's AI law, as even a vendor that sells ISO training states in its own guide. What it gives you is a management system that the Korean duties can be mapped onto, which is why it works as a baseline across jurisdictions.
The table is an orientation aid. ISO text is paywalled and not quoted here, so clause and control references are by title only.
| Korean duty | What the management system gives you | ISO 42001 hook (by title) |
|---|---|---|
| Accountability for the duties | Assigned and communicated roles and authority | Clause 5.3 Roles, responsibilities and authorities; Annex A.3.2 AI roles and responsibilities |
| Risk management for high-impact systems | A repeatable risk method and treatment plan | The AI risk assessment and treatment clauses |
| Impact assessment (Article 35, "endeavor") | A documented AI system impact assessment | The AI system impact assessment clause |
| Human oversight | Defined oversight of AI use | Annex A controls on the use of AI systems |
| Documentation and records | Controlled, retained evidence and management review | Documented information; clause 9.3 Management review |
| User notice | Information for interested parties | Annex A controls on information for interested parties |
| Suppliers and foreign providers | Third-party and supplier control | Annex A third-party controls |
| Concerns and incidents | A channel to raise AI concerns | Annex A.3.3 Reporting of concerns |
Watch out
A management system does not cover the representative designation, MSIT's confirmation and inspection process, the Korean labelling method or anything under PIPA. A certificate evidences a system. It does not answer a corrective order.
The cross-jurisdiction point is concrete. One inventory, one risk method and one record set can serve Korea, the EU and an enterprise buyer's security review. Our what is ISO 42001 guide explains the standard, and ISO 42001 sits alongside ISO 27001 rather than replacing it.
ComplyJet customers
A small AI company, audit-ready with a team guiding it
Raghuram Vadapally, Managing Director and Head of Software Engineering at Synexar AI, says ComplyJet has helped the team "stay organized and audit-ready at all times." Synexar, a company of 2 to 10 people, completed SOC 2 Type 1 and HIPAA reports in the same quarter. The engagement covered those two frameworks, not ISO 42001 or Korea.
Common Mistakes When Reading the South Korea AI Framework Act
These are easy to get wrong, because the law's structure is unusual.
- Treating the grace period as a pause on duties. The duties are in force. Fines, and generally investigations and corrective orders, are paused by policy, with serious-harm exceptions. Fix: build the records now.
- Assuming a labelling failure is fined directly. It draws a corrective order first. Fix: label, and keep proof of the method.
- Assuming "no Korean office" means "no Korean law." Article 4(1) looks at effect, not address. Fix: write the in-scope determination.
- Reading only the revenue threshold. The user-count test catches small companies with popular consumer products. Fix: check all three.
- Treating a listed sector as an automatic high-impact label. It is a two-part test with a prior review. Fix: run and date the determination.
- Copying the EU AI Act programme and stopping. Fines, the representative and the notice duty differ. Fix: map both laws to one record set.
- Treating ISO 42001 or any certificate as a safe harbour. It is a baseline. Fix: map Korean duties onto it and keep the Korea-specific steps.
- Relying on a 2025 summary. The Act has been amended and the decree amended since. Fix: check the date and the consolidated text before acting.
FAQs
What is the South Korea AI Framework Act?
It is South Korea's comprehensive AI statute, Act No. 20676, in force since 22 January 2026. It promotes the AI industry and sets duties for generative AI, high-impact AI and very large models. It reaches AI that affects Korean users or markets, and MSIT is the responsible ministry.
Is the South Korea AI Framework Act the same as the AI Basic Act?
Yes. They are two English renderings of one statute. The Korean title uses the word for "basic," so many law firms and government pages say AI Basic Act, while others say AI Framework Act. Both cite Act No. 20676 and the same obligations.
When did South Korea's AI law take effect?
The Act and its Enforcement Decree took effect on 22 January 2026. The Act was promulgated on 21 January 2025, with a one-year preparation period. MSIT paused administrative fines for at least a year, but the duties applied from the effective date.
Does the South Korea AI Framework Act apply to foreign companies?
Yes, where their conduct affects Korean users or the Korean market (Article 4(1)). Foreign operators with no Korean office must also designate a domestic representative if they meet a revenue or user threshold. How MSIT will apply the reach to companies with no Korean footprint is untested.
Does a US company need a domestic representative in South Korea?
Only if it has no domicile or place of business in Korea and meets one threshold: KRW 1 trillion total revenue, KRW 10 billion AI-service revenue, or one million average daily Korean users. It can also apply after a fine for ignoring a corrective order. Check the user test first.
What are the penalties under South Korea's AI Act?
Administrative fines are capped at KRW 30 million, about US$20,000 to US$21,000. Only three failures carry a direct fine: no advance notice, no domestic representative and ignoring a corrective order. Other failures draw an investigation and a corrective order first.
Is there a grace period on fines under South Korea's AI law?
Yes, as a ministry policy. MSIT said it will implement a grace period of at least one year before imposing administrative fines, and law firms report that investigations and corrective orders are also generally deferred outside serious-harm cases. It has no official end date and no extension has been announced.
Does ISO 42001 make a company compliant with the South Korea AI Framework Act?
No. ISO 42001 certifies a management system, not legal compliance. It gives you roles, risk and impact assessment, records and review that Korean duties can be mapped onto. The representative designation, labelling method and regulator process still need Korea-specific work and legal advice.
Book a demo
Build one AI governance programme for Korea, the EU and your customers
Talk through how to keep your AI policies, assessments and evidence audit-ready under ISO 42001, for a team your size. ComplyJet is compliance automation, so we do not give legal advice, act as a Korean representative or decide whether a system is high-impact.
Related Reading
- AI regulations by country: the multi-country map this article's South Korea section expands on
- EU AI Act compliance guide: the EU side of the comparison, with dates and obligations
- What is ISO 42001: the management-system standard used here as a baseline
Sources: the Korea Legislation Research Institute's English text of the Act (elaw.klri.re.kr), MSIT's English legislative-notice page, Cooley's 27 January 2026 overview, and the sources linked above. This article is general information, not legal advice.
