Every AI Policy Ever · open in the interactive register
Published (non-binding, current)Non-binding · Guidelines / framework
인공지능 투명성 확보 가이드라인
Record KR-04
Guidance from the Ministry of Science and ICT on the transparency duties of Article 31 of Korea's AI Basic Act: prior notice that a product or service is operated on the basis of high-impact or generative AI (para. 1); marking that outputs were generated by generative AI (para. 2); and notice or marking of deepfakes, defined as virtual sound, image or video outputs hard to distinguish from reality (text excluded), so that users clearly recognise them (para. 3). The duties fall on the operator that finally provides the product or service to users, including overseas operators affecting Korean nationals. It sets out notice channels and human-perceptible or machine-readable marking methods.
No English version of this instrument was found. The English title and the summary are the register's own, written from the sources below; for the exact wording use the Korean text.
Released by the Ministry of Science and ICT on 22 January 2026, when the Act entered into force, under the name 'Guidance on securing AI transparency (guideline)' (인공지능 투명성 확보 안내 지침(가이드라인)); a draft had been published in September 2025. NIA posted a revised file dated 26 January 2026 and recorded a wording change to the service examples for text outputs on 27 January 2026. Under the ministry's policy of a guidance period of at least one year, fact-finding and fines under the transparency provisions are suspended during that period, and the ministry plans to keep revising the guideline. Inside a service, outputs may be marked through the user interface, but outputs that can be downloaded or shared must carry the mark in the output itself; exported deepfakes need a mark that people can perceive intuitively, while artistic and creative works may use a lighter marking that does not hinder their display or enjoyment. Where a machine-readable mark is used, users must be told at least once by text or voice. A foundation-model developer that only supplies an API is not bound; the operator that offers the service to users is. The guideline coins the term 'general outputs' for AI outputs that are not deepfakes.
Open the official source https://www.nia.or.kr/common/board/Download.do?bcIdx=28987&cbIdx=99835&fileNo=14
Secondary source: https://www.korea.kr/briefing/pressReleaseView.do?newsId=156740669
Archived. A source changed since the previous snapshot; both versions are kept. Last attempt 2026-10-18 07:41:17 UTC
Korea AI Basic Act South Korea · In force
Korea AI Basic Act Enforcement Decree South Korea · In force
Korea frontier AI safety guideline (Art. 32, draft) South Korea · Draft / proposed
Korea high-impact AI determination guideline South Korea · Published (non-binding, current)
Ministry of Science and ICT (AI Safety and Trust Support Division); also published by the Telecommunications Technology Association (TTA) and the National Information Society Agency (NIA). Guideline on Securing AI Transparency (Article 31 of the AI Basic Act). 2026-01-22. https://www.nia.or.kr/common/board/Download.do?bcIdx=28987&cbIdx=99835&fileNo=14. Status: Published (non-binding, current); verified 2026-10-18 (level B). Local copy archive/KR-04/2026-10-12/official.pdf retrieved 2026-10-12 04:43:48 UTC, SHA-256 f3c9206fc9e8b33d1ac7cf973bb00422efd2b0296347fa0346456bd807ef020d. Every AI Policy Ever, version 1.18 (snapshot 2026-10-18), record KR-04.
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.