ITIF-一项法律制定了韩国的人工智能政策_一个薄弱环节可能会打破它（英）-2025.9_25页_474kb

South Korea's AI Framework Act Summary

Core Content

South Korea's AI Framework Act, passed in December 2024, is the world's first law to integrate strategy, industrial promotion, and regulation into a single statute. This comprehensive approach aims to coordinate government direction, promote AI development and adoption, and manage AI-related risks. However, the Act's regulatory provisions are criticized for being overly broad and potentially stifling innovation and competitiveness.

Main Viewpoints

• Unified Policy Instrument: The Act is designed as a single instrument to manage AI strategy, promotion, and regulation, offering both potential and risks.
• Strengths in Promotion: The Act's promotion agenda is strong, focusing on data infrastructure, clusters, talent, and internationalization, positioning South Korea as a leader in AI development.
• Regulatory Challenges: The regulatory section is seen as a major weakness due to its broad definitions, rigid mandates, and process-heavy requirements that may hinder innovation and scalability.
• Need for Balance: To achieve its goals, the Act must be adjusted both structurally and in terms of practical implementation, ensuring that it supports the AI ecosystem without imposing undue burdens.

Key Information

Chapter 1: General Provisions

• Defines core terms such as Artificial Intelligence, AI System, AI Technology, High-Impact AI, Generative AI, AI Industry, AI Business Operator, User, and Affected Person.
• Establishes the AI Society concept as a vision for AI's role in driving progress across all fields.
• Introduces AI Ethics as a guiding principle for responsible AI development and use.

Chapter 2: Governance for Sound AI Development and Trust

• Establishes the National AI Strategy Committee under the president, tasked with setting the national AI strategy and allocating resources.
• Requires the committee to approve the AI Master Plan, which outlines the government's strategic direction for AI.
• The AI Policy Center and AI Safety Research Institute (AISI) are created to provide technical support and evaluate AI risks.
• The law mandates that ministries must act on the committee's recommendations, potentially creating a master regulator with too broad authority.

Chapter 3: Policies for AI Development and Industry Promotion

• Focuses on innovation policies and industrial promotion.
• Includes provisions for funding R&D projects, promoting AI standards, and expanding AI training data infrastructure.
• Establishes an Integrated Provision System for managing training data, with free access for the private sector and potential fees for certain users.
• The Act promotes SMEs but may inadvertently stifle large-scale innovation by prioritizing them over bigger firms.

Chapter 4: AI Ethics and Trustworthiness

• Combines soft-law measures (ethical principles, voluntary ethics committees) with hard-law obligations (transparency, safety, impact assessments).
• Transparency obligations include mandatory labeling and disclosure, which are criticized as technically fragile and inconsistent across jurisdictions.
• Safety obligations are tied to compute thresholds, which are not reliable indicators of risk.
• The Act encourages performance-based oversight and targeted rules for specific harms like disinformation, IP violations, and online harassment.

Chapter 5: Supplementary Provisions

• Sets rules for funding and operating the Act's programs.
• Highlights the risk of regulatory overreach due to broad data demands, potentially affecting firms' proprietary information.

Chapter 6: Penalties

• Outlines penalties for non-compliance with key obligations.
• Suggests that penalties should be proportionate to the severity and impact of violations.
• Proposes a grace period for firms to adjust to new obligations before facing penalties.

Recommendations

• Amend Article 2 to narrow the definition of "AI system" to exclude ordinary software and focus only on systems that create novel governance challenges.
• Amend Articles 7–9 to remove the National AI Strategy Committee's authority to dictate regulatory changes, returning regulatory design to sector-specific ministries.
• Amend Article 13 to allow flexible R&D roadmaps instead of prescriptive mandates.
• Amend Article 17 to adopt size-neutral language in AI support policies.
• Amend Article 31 to replace mandatory disclosure requirements with voluntary provenance standards and targeted rules for specific harms.
• Amend Article 32 to remove compute thresholds as a regulatory trigger and instead rely on post-deployment evaluations.
• Amend Articles 33–35 to replace process-heavy self-assessments with performance-based requirements.
• Amend Article 36 to ensure equal treatment of domestic and foreign firms based on actual risk, not revenue or user size.
• Clarify Enforcement Decrees for Article 14 to ensure industry-led standards, Article 18–19 to maintain size-neutral support, and Article 40 to limit data requests to what is necessary and protect confidentiality.
• Scale penalties proportionally and introduce a grace period for compliance.

Conclusion

The AI Framework Act has the potential to be a powerful tool for shaping South Korea's AI future. However, to realize this potential, the law must be refined to address structural flaws and ensure balanced, practical implementation. By focusing on strategic unity while allowing sector-specific regulation, and by promoting flexibility and performance-based oversight, South Korea can better support its AI industry and maintain global competitiveness.