Polity Atlas
News · AI & the Future of GovernanceEstablished

South Korea's AI Basic Act takes effect

Korea's Framework Act on the Development of Artificial Intelligence and the Creation of a Foundation for Trust took effect on 22 January 2026. It defines 'high-impact' AI, including state decisions on eligibility for public services, requires labelling of generative AI output and sets fines of up to 30 million won.

Polity Atlas editorsPublished 22 Jan 2026
Original sourceKorea Legislation Research Institute (KLRI), Statutes of the Republic of KoreaRead at Korea Legislation Research Institute (KLRI), Statutes of the Republic of Korea ↗

What happened. The National Assembly passed the Act in December 2024; it was promulgated on 21 January 2025 as Act No. 20676 and entered into force one year later, on 22 January 2026 (KLRI English text). It was amended once before taking effect, by Act No. 21311 promulgated on 20 January 2026 (consolidated text). An Enforcement Decree sets the detailed thresholds; its current text is Presidential Decree No. 36580, in force since 20 August 2026 (Korea Law Information Center). The Ministry of Science and ICT, the competent authority, has said it will grant a one-year grace period before administrative fines are imposed (Cooley).

What it does.

  • High-impact AI (Article 2) covers systems that may significantly affect life, physical safety or fundamental rights in listed areas, including energy and drinking water, health care, nuclear facilities, biometric analysis in criminal investigations, hiring and loan screening, transport, student evaluation, and decisions by the state, local governments and public institutions that affect citizens, such as verifying qualifications for public services or collecting charges.
  • Duties for operators of high-impact systems (Article 34) include risk management, an explanation plan covering outputs, main decision criteria and training data, user protection, human oversight and record-keeping. Prior assessment of impacts on fundamental rights is an obligation to 'endeavour' (Article 35), but state agencies and other public entities that intend to use high-impact products must give preferential consideration to those that have been assessed.
  • Transparency (Article 31): users must be told in advance when a high-impact or generative AI service is used, generative outputs must be labelled, and realistic synthetic sound, images or video must be clearly marked.
  • Frontier models above a compute threshold set by the Enforcement Decree at 10^26 floating-point operations of cumulative training compute carry lifecycle risk-assessment and incident duties (Article 32).
  • Foreign operators above user or revenue thresholds must appoint a domestic agent (Article 36).
  • Governance: a Presidential Council on National AI Strategy of up to 60 members, chaired by the President (Articles 7-8, as amended).
  • Penalties: administrative fines of up to 30 million won (Article 43) for failing to give notice, to appoint an agent, or to comply with corrective orders.

Why it matters. Korea is among the first states anywhere with a horizontal AI statute in force, and it sits between the EU's risk regulation and Japan's promotion law. By listing state eligibility decisions as high-impact, it brings administrative AI within its scope, but the public-sector lever is mainly procurement preference rather than a direct duty to assess.

What to watch. Use of corrective orders after the grace period; whether the impact-assessment preference actually shapes government procurement; and further amendments, since the Act was already amended once before taking effect.