South Korea weighs jail terms for tampering with AI watermarks

South Korean lawmakers are considering a proposal from Rep. Kim Daesik that would require AI-generated material to carry a watermark embedded in the output itself and make tampering with that marker a criminal offence. The measure remained before the National Assembly as of Oct. 17, alongside other unresolved efforts to clarify AI-content labeling rules. Kim, identified as Kim Dai-sik in earlier English-language reporting, introduced the bill with nine other People Power Party legislators. It would amend the AI Basic Act, formally the Framework Act on the Development of AI and Establishment of a Foundation of Trust, by requiring codes, text or symbols to be placed in the generated file rather than only in a platform interface. The proposed approach is intended to keep an AI-origin notice attached when content is downloaded, edited or reposted. Existing interface captions, icons or banners can disappear when secondary creators screenshot, crop or repackage material, leaving downstream removal outside a clearly enforceable prohibition, according to the Korea Times. Damaging, forging or altering a required watermark would carry a maximum penalty of two years in prison or a fine of 20 million won, about $13,500, under Kim’s bill. The lawmakers argued that the present gap allows AI-generated material to circulate as though it had been made by a person or represented real events. The AI Basic Act and its enforcement decree took effect Jan. 22. Article 31 requires AI businesses providing generative AI, or products and services using it, to indicate that an output was AI-generated. For exported material such as downloaded or shared content, the label must be applied directly to the output, while artistic or creative deepfake works may use notices that do not interfere with viewing or enjoyment. Yet the current framework provides no legal basis to punish a party other than an AI business operator for removing watermarks or markings during distribution. It also draws a distinction between AI developers and AI-using business operators, while treating people who simply use generated output to make their own content as users rather than regulated AI businesses. The Assembly’s Legislation Review Subcommittee discussed the related amendments on Sept. 15 but deferred a decision, saying further consideration was needed. The broader legislative debate includes proposals to allocate machine-readable and user-perceptible labeling responsibilities by the roles of developers and AI-using businesses, exempt limited assistive uses such as background generation or color adjustment, and require platforms and creators to take anti-tampering measures. The unresolved issue is not only where a label sits, but when it is required and who must apply it. Lee Yongmin, a lawyer at Yulchon, said rules must specify how extensively AI was used at each production stage, including where part of a work comes from an outside contractor. Park Hyungjun of Sungkyunkwan University said continuously reworked AI content makes it difficult to decide whether excerpts reused in another work still require labeling, and argued that legal requirements should be supplemented by self-regulatory guidance. Kim’s bill is separate from the Ministry of Science and ICT’s planned extension of deepfake-labeling duties to users and distributors, including YouTubers, as we reported Oct. 8. That ministry proposal remains under discussion, including the scope of users who would be covered.