Korea's framework AI act takes effect
The Framework Act on the Development of Artificial Intelligence and Establishment of Trust came into force: labelling of generative AI output, explanations and user protection plans for high-impact AI, and safety duties for the most powerful systems, whose threshold, set by decree, begins at 10^26 operations of training compute.
Why it matters
Korea's ministry calls it the world's second framework AI act; the first was the EU's. A compute figure became a binding trigger, but through the enforcement decree rather than the statute, and only as one of three criteria that must all be met.
The act was passed on 26 December 2024, promulgated on 21 January 2025 as Act No. 20676 and took effect a year later, on 22 January 2026. Operators must tell users in advance that a product or service runs on high-impact or generative AI, label generative AI output, and clearly signal sound, images or video that are hard to distinguish from the real thing. Operators of high-impact AI must explain final results as far as technically feasible and run user protection plans. The act leaves the threshold for the most powerful systems to a presidential decree, which sets cumulative training compute of 10^26 floating-point operations or more, together with state-of-the-art technology and a risk of broad, serious impact on life, safety and fundamental rights; such systems must identify, assess and mitigate risks and run a risk management system. Foreign companies without an office in Korea must designate a domestic representative if they meet any of the decree's thresholds: a trillion won of revenue, ten billion won from AI services, or a million daily users on average. Fines of up to 30 million won apply only to failing to give the advance notice, to designate a representative, or to obey a stop or correction order. The ministry announced a grace period of at least a year, during which it neither investigates nor fines.