The Ministry of Culture of South Korea and the national copyright regulator have issued new guidance on how copyright law and fair use provisions may apply to the training of generative artificial intelligence models. The move occurs as governments around the world face pressure to tackle the intersection of technological innovation and the rights of content creators.

Countries are increasingly examining how IP frameworks should address generative AI, which relies on large datasets to spot patterns and generate new content. South Korean officials acknowledge that while AI can fuel economic growth, the unauthorized use of creative works has become a flash point for authors and rights holders.

According to the framework outlined in the guide, training generative AI models may involve copyright-relevant acts because works are often reproduced during data collection, preprocessing and model training. In general, using copyrighted works for AI training requires the rightsholder’s prior permission, unless an exception such as fair use applies. The guide stresses that fair use is evaluated on a case-by-case basis and is ultimately determined by the courts.

The document identifies four key factors for assessing fair use in the context of AI: the purpose and character of the use, the nature of the copyrighted work, the amount used and the effect on the potential market. Uses that are innovative and serve public purposes, such as models for research or analysis, may be more likely to be considered fair use, especially if they do not reproduce original content or disrupt existing markets. Commercial uses that closely replicate or substitute for original works are less likely to qualify.

The report also points to a shift towards structured licensing markets for AI training data. Companies and rights holders are increasingly entering into agreements covering the use of content in AI systems, including deals in publishing, music and image licensing. These arrangements range from revenue-sharing models to large-scale content licenses, suggesting that AI data sourcing is becoming more formalized.

The differences between jurisdictions remain a defining feature of the global intellectual property landscape. While countries such as the United States and South Korea rely on flexible fair use doctrines, others, including the European Union and Japan, have adopted special “text and data mining” exceptions that permit certain uses of copyrighted works under defined conditions. These divergent approaches have created legal uncertainty for technology companies operating across borders and contributed to ongoing litigation.

Recent court cases cited in the guide illustrate the unresolved state of the law. In the United States, a court suggested that training AI models on lawfully purchased books could constitute fair use, while a German court reached a different conclusion under local law.  The guide warns that many such decisions are preliminary and may be revised on appeal, noting that statutory standards are still evolving.  The document also points to a growing awareness that AI-generated output may raise intellectual property issues if it closely resembles the underlying works. Outputs substantially similar to human-created originals could lead to infringement claims and disputes between developers, users and rightsholders.

To manage these tensions, the South Korean authorities have proposed plans to support licensing negotiations, improve access to rights information and increase the availability of public-sector data for AI training. New licensing categories introduced in 2026 allow certain government-owned works to be freely used for AI purposes under designated conditions, with the aim of giving developers legally clear data sources.

Collectively, the guidance points to several trends driving the intellectual property environment in 2026: the growth of structured data licensing, the continued reliance on flexible but evolving legal doctrines such as fair use, divergence among national regulatory models and a rising volume of cross-border litigation. It reflects an approach that seeks to advance both AI innovation and copyright protection, rather than treating them as competing interests.

- Cathy Li