South Korea ranks second worldwide in paid ChatGPT subscribers, but critics say the country lags others in protecting copyright in the age of artificial intelligence. Europe and Japan have already built regulatory frameworks, while Korea has only begun discussions — and those have focused more on AI-generated works than on the use of copyrighted material to train AI.
OpenAI's ChatGPT had 17,993,884 monthly active users in Korea in September, the most of any AI app, according to Mobile Index by Korean data firm IGAWorks. That was up 40.6% from September last year. Anthropic's Claude ranked second with 1,708,413 users, a 1,211.3% jump from 130,286 a year earlier — more than a 13-fold increase in a single year. Globally, Korea accounted for 5.4% of ChatGPT revenue by country, the second-highest share after the United States at 35.4%, according to market research firm Sensor Tower.
Yet the legal framework for AI and copyright remains underdeveloped. Korea's AI Framework Act took effect in January, but critics say it lacks explicit provisions to protect works created by humans and that its labeling requirements for AI-generated content are vague.
Attention has centered less on AI training than on low-quality "one-click publishing" — books produced almost entirely by AI, such as translations run through an AI model and published as-is, with little involvement from an author or translator. With the arrival of AI publishing, some expect the number of new titles released each year to swell to between 200,000 and 300,000, from an annual average of about 62,000 over the past five years. Before long, books written by AI could outnumber those written by people. The National Assembly and the industry have moved to respond. In March, lawmakers cleared the way for libraries to decline legal deposit copies of AI-produced publications. The Korean Publishers Society held a policy forum last month and announced a principle calling on publishers to clearly disclose the extent of their use of AI.
Discussions on AI training are also under way. At a meeting on AI and copyright held on the 8th, the Presidential Council on Intellectual Property, which reports to the prime minister, reviewed and approved measures on the use of copyrighted works to train AI. The core provision would require prior permission from rights holders when copyrighted works are stored or processed during AI training. In February, the Ministry of Culture, Sports and Tourism issued a guidebook setting out criteria for judging fair use in the use of AI training data, and said it would open an AI-focused counseling and dispute mediation channel at the Korea Copyright Commission. The ministry plans to distribute a standard contract this year covering the transfer and use of copyrighted works for AI training purposes. Still, cases like Anthropic's suggest that more assertive action is needed to protect Korean authors' publications against large global AI companies.
The publishing industry says it must also prepare carefully for situations unimaginable under the current copyright law. "Plagiarism by humans used to be relatively clear-cut, but with AI, it has become far more subtle and complicated," one industry official said. "For example, if the content is entirely different but the writing style is identical to that of a real, living author, it is hard to identify and hard to punish." If an AI trained on fictional works confuses the fabricated with the real and gives users false information, the question of who bears responsibility could also become contentious.