South Korea plans to make AI users label deepfakes
A proposed amendment would shift watermarking duties toward creators and distributors, including YouTubers
South Korea plans to shift part of the responsibility for labeling deepfakes from AI companies to the people who create or distribute the material. The proposed amendment to the country’s AI Framework Act would require users, including YouTubers, to add watermarks to deepfake content themselves.
The Ministry of Science and ICT is drafting the change to the Framework Act on the Development of AI and Establishment of a Foundation of Trust, Seoul Economic Daily reported. The scope of users who would be covered remains under discussion, meaning the proposal has yet to define the full reach of the new duty.
Current rules place the obligation on AI companies to disclose visibly and clearly when material is a deepfake. The law defines such content as virtual sound, images or video that are difficult to distinguish from reality. It permits invisible labels for other AI-generated material, while reserving visible labels for deepfakes.
That distinction has created a technical problem for providers. AI systems cannot reliably judge whether their own output will appear real to a person and therefore qualify as a deepfake. Companies have consequently been forced to make all labels visible to ensure compliance, according to a senior government official quoted by the newspaper.
Officials are seeking to put the watermarking task on users who produce or distribute deepfake material while retaining the labeling regime. The government has suspended fines for violations of the existing requirement until at least Jan. 22, 2027, to limit disruption to industry, but officials see little room to extend the watermarking duty beyond January as misuse spreads into voice phishing, sex offenses and the content industry.
The underlying AI Framework Act was enacted on Jan. 21, 2025, and took effect a year later. Its stated aims include protecting human rights and dignity, regulating the sound development of AI and establishing trust. It applies as well to actions abroad that affect South Korea’s domestic market or users.
South Korea criminalized the consumption of deepfake pornography in September 2024. Police handled 812 reports of deepfake-related sex offenses from January through September that year, compared with 156 reports in all of 2021. The earlier amendments set penalties of up to three years in prison or a 30 million won fine for creating, possessing or viewing such material.