Framework Act on the Development of Artificial Intelligence and the Creation of a Foundation for Trust

Summary

AI business operators must tell users in advance that a product or service runs on high-impact or generative AI, mark generative AI outputs, and give notice of or mark outputs hard to tell from reality (Art. 31). Operators of AI systems trained with cumulative compute at or above a decree threshold must identify, assess and mitigate risks over the life cycle, monitor and respond to safety incidents and report to the Minister (Art. 32). For high-impact AI, operators must review in advance whether their AI qualifies (Art. 33), take risk management, explanation, user protection and human oversight measures (Art. 34) and endeavour to assess impacts on fundamental rights (Art. 35).

Language

English translation by Korea Legislation Research Institute (KLRI) Open it. The Korean text is authoritative.

Safety topics

Open question

checker: The evidence script's error on date_in_force is resolved: 2026-01-22 is correct (Addenda: in force on the day after one year from promulgation on 21 Jan 2025; the MSIT guideline stored for KR-06 states 'in force from 22 Jan 2026', the stored decree text in KR-03 also takes effect on 22 Jan Issue #14

The record shows its last checked values until a maintainer decides; the decision is recorded in its history.

Status notes

Promulgated on 21 Jan 2025 as Act No. 20676 and in force one year later, on 22 Jan 2026 (Addenda Art. 1); the digital medical device part of Art. 2(4)(d) applies from 24 Jan 2026. Act No. 21311 of 20 Jan 2026 amended it from 22 Jan 2026, renaming the National AI Committee the National AI Strategy Committee (Art. 7); its provisions on AI-vulnerable groups (Art. 3(5)), support for AI use costs (Art. 17-2), a state AI research body (Art. 22-3) and reflecting AI-vulnerable groups in impact assessments (Art. 35(1)) took effect six months after promulgation, on 21 Jul 2026. Art. 12 provides for the AI Safety Institute (KR-01). The Act applies to acts abroad that affect the Korean market or users and excludes decree-designated AI used only for defence or national security (Art. 4). Foreign operators above decree thresholds must designate a domestic agent (Art. 36). Enforcement powers include fact-finding investigations and stop or correction orders (Art. 40); administrative fines of up to KRW 30 million apply for failing to give the prior notice of Art. 31(1), failing to designate a domestic agent, or ignoring such orders (Art. 43).

Sources

Open the official source https://www.law.go.kr/법령/인공지능발전과신뢰기반조성등에관한기본법

Secondary source: https://elaw.klri.re.kr/eng_service/lawView.do?hseq=71019&lang=ENG

Local copy

Verification

History

1. 2026-10-14verification.level: B → A — level computed from the evidence by tools/pipeline.py2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)
2. 2026-10-14english_version: {"kind": "government translation", "url": "https://elaw.klri.re.kr/eng_service/l → {"kind": "government translation", "url": "https://elaw.klri.re.kr/eng_service/l — Points to KLRI's English text frame, now stored, instead of the empty viewer page.2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)
3. 2026-10-14notes: Passed by the National Assembly on 26 Dec 2024, consolidating 19 bills (per OECD → Promulgated on 21 Jan 2025 as Act No. 20676 and in force one year later, on 22 J — Added Act No. 21311 with its staggered commencement, Art. 4 (scope), Art. 12 and Art. 36. Removed the passage date and the 19 consolidated bills attributed to OECD.AI (no stored file supports them; now in verification.note). The Art. 43 fine now names the Art. 31(1) prior notice.2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)
4. 2026-10-14safety_topics: ["risk-based classification", "general-purpose / frontier AI", "transparency & l → ["risk-based classification", "general-purpose / frontier AI", "transparency & l — Cut from 8 to the 6 the vocabulary allows, keeping the topics the Act's main duties support; dropped accountability & liability and incident reporting.2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)
5. 2026-10-14summary: Comprehensive AI framework law with binding obligations. Operators of high-impac → AI business operators must tell users in advance that a product or service runs — Rewritten from the Korean text. Art. 33 is a prior self-review, with confirmation by the Minister only on request (not 'must confirm'); Art. 35 is an effort duty called 영향평가 (impact assessment), not a 'fundamental-rights impact assessment'; Art. 31(3) applies to any AI business operator and allows notice or marking; the Art. 32 threshold is 'at or above'; 'operators' is now 'AI business operators' (인공지능사업자). The domestic agent and the AI Safety Institute moved to notes.2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)
6. 2026-10-14date_last_updated: – → 2026-01-20 — Last amended by Act No. 21311 of 20 Jan 2026 (partial amendment); the current version is in force from 21 Jul 2026.2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)
7. 2026-10-14english_version: – → {"kind": "government translation", "url": "https://elaw.klri.re.kr/eng_service/l — English version found by the translation check in a stored copy (archive/KR-02/2026-10-11/secondary.html).manual-2026-10-14 · Claude Code (claude-opus-5-5)
8. 2026-10-14language: Korean (English reference translation by KLRI) → Korean — The translation check found the official text in this language only; the English texts are translations (recorded in english_version where stored).manual-2026-10-14 · Claude Code (claude-opus-5-5)

Related records

Korea AISI South Korea · Published (non-binding, current)

Korea AI Basic Act Enforcement Decree South Korea · In force

Korea AI transparency guideline (Art. 31) South Korea · Published (non-binding, current)

Korea frontier AI safety guideline (Art. 32, draft) South Korea · Draft / proposed

Korea high-impact AI determination guideline South Korea · Published (non-binding, current)

Korea high-impact AI operator duties guideline South Korea · Published (non-binding, current)

Korea AI impact assessment guideline South Korea · Published (non-binding, current)

Citation

National Assembly of the Republic of Korea (competent ministry: Ministry of Science and ICT). Framework Act on the Development of Artificial Intelligence and the Creation of a Foundation for Trust [Act No. 20676]. 2025-01-21. https://www.law.go.kr/법령/인공지능발전과신뢰기반조성등에관한기본법. Status: In force; verified 2026-10-14 (level A). Local copy archive/KR-02/2026-10-11/official.html retrieved 2026-10-11 14:02:44 UTC, SHA-256 274881056bc7039f627a2491d078255555ad912507b5f20171d6cc4d1807beb0. Every AI Policy Ever, version 1.18 (snapshot 2026-10-18), record KR-02.

Improve this record

Know of an amendment, a status change or an error? Report news or an error on GitHub; the form opens with this record filled in.