South Korea’s AI Basic Act, now in force, establishes the most comprehensive transparency requirements for AI operators, with extraterritorial reach and enforcement measures that could influence global AI governance standards.
South Korea has moved into the front rank of AI regulation with a framework that makes transparency a legal duty rather than a voluntary best practice. The country’s AI Basic Act, which took effect on 22 January 2026, is the first comprehensive AI statute in Asia and places Article 31 at the centre of its disclosure regime. According to the official text and the Ministry of Science and ICT’s guidance, the provision is designed to make users aware when they are dealing with high-impact or generative AI, and when content has been created or altered by AI systems.
The obligations do not fall on every end user of AI. They are aimed at AI business operators: the developers and providers of AI services, and the companies that build products on top of model outputs for external users. The distinction matters in practice. A company that embeds a generative tool into a customer-facing service may be subject to the disclosure rules, whereas a studio or individual simply using AI-generated material internally will usually be treated differently. The law also reaches beyond South Korea’s borders if a service affects the Korean market or Korean users, which gives Article 31 an extraterritorial reach that foreign software and outsourcing firms cannot ignore.
Article 31 sets out three separate layers of disclosure. First, users must be told in advance when a product or service relies on high-impact or generative AI. Secondly, generative outputs must be labelled as such. Thirdly, content that is hard to distinguish from reality, including realistic audio, image and video deepfakes, must be clearly identified in a way users can immediately recognise. The Library of Congress and the official Korean legal text both describe these duties as core transparency obligations under the Act.
The Ministry of Science and ICT’s guidance, issued on the day the law came into force, gives the clearest operational detail. Prior notice may be delivered through terms of service, an on-screen banner, onboarding text or a visible notice at the point of use. For outputs that remain within a service, a logo, caption or spoken notice may be sufficient, depending on the format. But where content can be downloaded or shared outside the platform, the label must travel with the output itself through visible markings, metadata or a watermark. The guidance draws a sharper line for deepfakes: machine-readable markers alone are not enough, because the user must be able to see, hear or otherwise recognise that the material is AI-generated.
There are limited exceptions. Where AI use is obvious from the service itself, where the system is used purely for internal business purposes, or where creative works would be distorted by ordinary labelling, the method of notice can be adapted. That flexibility does not remove the duty; it only changes how it is delivered. The law also provides for enforcement measures and fines of up to KRW 30 million, although the ministry has announced a grace period of at least one year from commencement. In practice, that means penalties are not expected before January 2027, even though the compliance duties already apply.
For foreign providers, the compliance picture is more demanding still. The Act requires some overseas operators to appoint a local representative in South Korea if they meet threshold tests based on revenue or user numbers. That requirement, together with the transparency rules, means that companies serving Korean users need to map where AI is used in their products, decide whether outputs stay inside the service or are exported, and document the technical measures they rely on. The practical lesson is straightforward: disclosure is now part of product design, not an afterthought, and firms that build the notices and labels into their interfaces early will be better placed to meet Korean requirements as well as similar rules emerging elsewhere, including under the European AI Act.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
