Korea's groundbreaking AI law requires watermarks on generated content, but enforcement gaps remain

Korea on Thursday began enforcing the world’s first comprehensive law governing artificial intelligence (AI), requiring watermarks on images, videos and audio created and distributed using generative AI.

  But concerns remain over whether the requirement can be enforced in practice, as watermark-removal applications remain widely available. Deepfake content produced using foreign tools also largely falls outside the law’s scope, raising fears that domestic AI companies could face heavier regulatory burdens than overseas rivals.

The AI Basic Act — officially titled the Basic Act on the Development of Artificial Intelligence and the Establishment of a Foundation for Trustworthiness — outlines how businesses must label AI-generated content under new transparency rules. 

  For content such as animation or webcomics that viewers can more easily identify as artificial, the law allows the use of invisible digital watermarks. Deepfake content that closely resembles real people or events must display a clear and visible watermark. The rules apply only to businesses, not individual users.

  Enforcement remains a key challenge. A large share of deepfake content circulating online is created using overseas apps that do not fall under the law’s jurisdiction. 

  While the legislation includes a provision requiring foreign technology companies to appoint a local representative in Korea, it applies only to firms that meet at least one of three thresholds: global revenue of 1 trillion won ($680 million), domestic revenue of 10 billion won or an average of 1 million daily users in Korea. In practice, that limits the measure to Google and OpenAI.

 

Watermarks themselves are also easy to remove. An overseas image-editing app that has been downloaded more than 500,000 times from Google Play openly advertises an “AI eraser” function designed to delete watermarks embedded across images. User reviews say the feature can remove watermarks with just a few clicks.

  The law has also done little to address the spread of so-called AI slop, a term used to describe low-quality or deceptive content mass-produced using generative AI. According to video-editing platform Kapwing, 11 Korea-based AI slop channels ranked among the top 100 YouTube channels by total views worldwide, with a combined 8.45 billion views, the highest among countries surveyed.

  "Regulating platforms is key to dealing with AI slop, but concerns about friction with the U.S. government appear to have kept the AI Basic Act from going that far," said Yoo Hyun-jae, a professor of journalism and mass communication at Sogang University.

 

Industry groups have voiced unease that the law could disproportionately affect domestic firms. 

  "The structure ultimately places additional burdens only on Korean companies," one industry official told the JoongAng Ilbo.

  Just 2 percent of 101 Korean AI startups had established a substantive compliance system to respond to the new law, according to a survey released last month by the nonprofit Startup Alliance.

  "Compared with Europe’s AI law, this is not regulation-only, but Korean companies competing with less-regulated U.S. and Chinese firms may still see it as reverse discrimination," said Wi Jong-hyun, a professor at Chung-Ang University’s College of Virtual Convergence.

  The government has sought to reassure industry. Authorities said they would not impose fines during a grace period of up to one year as the law takes effect. 

  "This law is based on the principle of minimal regulation and places emphasis on promoting the AI industry,” said Shim Zee-seop, a senior official at the Ministry of Science and ICT’s Artificial Intelligence Safety and Trustworthiness Policy Division. "We will ensure that AI innovation is not undermined.”

This article was originally written in Korean and translated by a bilingual reporter with the help of generative AI tools. It was then edited by a native English-speaking editor. All AI-assisted translations are reviewed and refined by our newsroom.
BY LEE YOUNG-KEUN [[email protected]]