Loading…
Loading…
The Framework Act on the Development of Artificial Intelligence and the Establishment of a Foundation for Trustworthiness — universally referred to as the AI Basic Act — consolidates 19 competing AI bills tabled in Korea's 22nd National Assembly into a single horizontal statute. It is the second comprehensive AI law in the world (after Regulation (EU) 2024/1689) and the first in Asia-Pacific.
Key structural features:
"High-impact AI" (고영향 인공지능) is defined as systems whose use may significantly affect human life, safety, or fundamental rights — explicitly enumerated to include healthcare diagnostics, energy/utility infrastructure, public services, criminal-justice decisions, recruitment / HR / promotion / disciplinary tools, credit-scoring, and education-grading. High-impact systems carry documentation, risk-assessment, transparency, and human-oversight duties.
label outputs that are wholly or substantially AI-generated; deepfakes and AI-generated content depicting persons must be conspicuously disclosed.
outside Korea where the effect is felt in the Korean market or on Korean users. Foreign providers without a Korean place of business above a threshold must designate a local representative responsible for compliance — directly mirroring the EU AI Act Article 22 and GDPR Article 27 representative-designation patterns.
Institute under MSIT for evaluation, red-teaming, and technical-standard work; (ii) a National AI Committee** chaired by the President as the cross-government policy "control tower"; (iii) standardisation, certification, and conformity-assessment authorities.
domestic AI: data-centre build, training-data access programmes, SME / startup standardisation support, R&D funding, and a national AI strategy.
(~USD 21,000) per violation plus targeted criminal liability (deepfake provisions). A one-year grace period before full penalty enforcement is government-announced. Fines are an order of magnitude smaller than the EU AI Act's EUR 35m / 7% global turnover ceiling — the IAPP characterised the Korean regime as "all roar, no bite" in monetary terms, though reputational and market-access effects remain material.
Subordinate regulations. MSIT issued a draft Enforcement Decree on 8 September 2025 and a package of supplementing regulations covering high-impact-AI designation criteria, generative labelling specifics, representative-designation thresholds, and sandbox / certification procedures. These were finalised through late 2025 ahead of the 22 January 2026 commencement.
is ~3 orders of magnitude smaller than the EU regime, and the market is one-tenth the size. The economic chilling effect on global AI providers is meaningfully lower than the EU comparator.
to enact comprehensive horizontal AI law, has explicit extraterritorial reach with mandatory representative designation, and operates one of the densest AI-using corporate ecosystems globally (Samsung, LG, SK Hynix, Naver, Kakao, Coupang). Compliance costs for US hyperscalers and global SaaS providers serving Korean users are non-trivial. The Act also signals to other Asia-Pacific jurisdictions (Japan, Singapore, Australia) that horizontal AI law is becoming a regional baseline.
(035720.KS), Samsung Electronics (005930.KS) gain regulatory clarity and explicit state R&D / data-centre support — bullish read for the EWY ETF basket.
GCP, OpenAI, Anthropic, Meta serving Korean users must designate local representatives, document high-impact systems, and meet generative-AI labelling. Marginal opex; not a market exit risk given the modest fine ceiling.
template-setter. Japan's "AI Promotion Act" (passed May 2025) and ongoing Singapore Model AI Governance Framework v2 iterations are likely to draw on the Korean text — particularly the high-impact taxonomy and representative-designation pattern. Watch for ASEAN convergence in 2026–2027.
Korea K-Chips Act (2023-03-31), Korea outbound investment screening (2024-11-15), and broader MSIT digital-strategy envelope. Together these constitute Korea's response to both the US-China tech decoupling and the EU regulatory wave.
Anthropic, xAI) as high-impact under the Enforcement Decree?** Threshold language remains under interpretation.
Information Protection Act already constrains training-data pipelines; the AI Basic Act adds layered duties — guidance on the interface is pending.
content.** Korea's election-period deepfake provisions (separately codified in the Public Official Election Act) intersect with the Basic Act labelling regime; authority allocation between NEC and MSIT is unsettled.