Korean bill seeks strict watermark mandate on AI-generated content - The Korea Times

Go to Top

Business<br>Tech & Science

Korean bill seeks strict watermark mandate on AI-generated content<br>ListenListen<br>Text Size<br>Print

By Jung Min-ho Published May 17, 2026 1:58 pm KST<br>Updated May 18, 2026 1:53 pm KST

Science and ICT Minister Bae Kyung-hoon listens to an explanation of a “deepfake detection and AI agent safety evaluation” during his visit to the Korea Artificial Intelligence Safety Institute in Seongnam, Gyeonggi Province, Oct. 23, 2025. Courtesy of Ministry of Science and ICT

Lawmakers are seeking to tighten Korea’s new artificial intelligence (AI) rules with a bill that would mandate watermarks on AI-generated content and criminalize their removal, in a bid to close what they call a “blind spot” in transparency rules.<br>The bill, introduced last week by Rep. Kim Dai-sik and nine other legislators of the main opposition People Power Party, seeks to amend the AI Basic Law (officially, the Framework Act on the Development of AI and the Creation of a Foundation for Trust), which took effect earlier this year.<br>The current law requires AI service providers to notify users whenever highly realistic audio, images or video have been created by an AI system. However, it does not specify where that notice should be placed or what form it should take, so most services satisfy the rule only with a small caption or icon inside their own interface. But once secondary creators screenshot, crop or repackage that content, the label is often removed — and under the current law, those downstream actors are not clearly treated as violators for removing it.<br>This gap allows AI-generated content to circulate online “as if it were created by a human or depicted real events,” the lawmakers wrote in the proposal.<br>The bill is designed to close that gap by shifting the focus from the platform’s own user interface — like a caption, banner or icon — to watermarks embedded in the output file itself. It would require providers to embed “codes, letters or symbols” into the content itself. The aim is to ensure that information about an AI origin stays attached to the file as it is edited, downloaded and reposted, and to make it harder for secondary creators to pass AI-generated material off as real by simply stripping away on‑screen labels.<br>The bill also targets those who tamper with these marks, prohibiting anyone from damaging, forging or altering the required notices or watermarks. Violations would be punishable by up to two years in prison or a fine of up to 20 million won ($13,500).