South Korea: If the AI product competes directly with the original market (or acts like a drop-in replacement), fair use becomes much harder to defend.
Summary
South Korea's Ministry of Culture, Sports and Tourism, in conjunction with the Korea Copyright Commission, released a guidance document in February 2026 clarifying when training generative AI on copyrighted works may constitute "fair use" and when it likely infringes. This document, while not legally binding, serves as a critical regulatory signal and compliance roadmap for AI developers, rights owners, and courts. It explicitly details how AI training involves reproduction and transformation of works, making fair use the central balancing mechanism assessed on a case-by-case basis. The guidance emphasizes "transformative use" as a key factor, indicating that extracting patterns is more defensible than creating AI products that directly compete with or substitute original copyrighted works. Crucially, it draws a sharp line on unlawful data access, bypassing access controls, circumventing technical protection measures, or ignoring robots exclusion signals, which can undermine a fair use defense.
Key takeaway
For AI developers building commercial services, your product design choices and data sourcing practices are paramount for copyright compliance. Focus on creating AI that extracts patterns rather than directly substituting existing markets, and ensure all training data is acquired legitimately, respecting access controls and technical protection measures. Documenting data provenance and compliance efforts will be crucial for defending against infringement claims, as regulators increasingly scrutinize how data is obtained and utilized.
Key insights
Fair use in AI training hinges on transformative use and legitimate data acquisition, not market substitution.
Principles
- Fair use is assessed case-by-case.
- Transformative use extracts patterns, not substitutes.
- Unlawful data access negates fair use.
Method
Assess AI training for fair use by evaluating if it extracts non-expressive patterns, avoids market substitution, and respects access controls and technical protection measures during data collection.
In practice
- Document data provenance rigorously.
- Implement model controls to reduce output infringement.
- Respect robots.txt and access restrictions.
Topics
- AI Copyright Law
- Fair Use Doctrine
- Generative AI Training
- Data Provenance
- Regulatory Frameworks
Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, Legal Professional, Policy Maker
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by Pascal’s Substack.