Seoul Sets The Pace: South Korea Enforces The World’s First Comprehensive AI Safety Law

Korea officially activates the world’s first enforced comprehensive AI regulatory framework, outpacing the European Union

As of January 22, 2026, South Korea has officially become the first country in the world to enforce a comprehensive AI regulatory framework. While the European Union (EU) passed its "AI Act" earlier, South Korea is the first to move from legislation to active, total enforcement across its entire national market.

Known formally as the "AI Basic Act" (or the Act on the Promotion of Artificial Intelligence Development and the Establishment of a Trust-Based Foundation), this law is a unique "hybrid" model. The law promotes massive industrial growth while setting strict guardrails for safety.

The "High-Impact" Risk Model

The law doesn't regulate all AI equally. It focuses on High-Impact AI, which includes systems that directly affect human life, safety, or fundamental rights.

Sector	Examples of High-Impact AI
Finance	Credit scoring and loan approval algorithms.
Employment	AI is used for resume screening or performance evaluation.
Healthcare	Diagnostic AI and medical advice tools.
Infrastructure	Management of energy grids, nuclear facilities, and water.
Public Safety	Biometric identification and criminal investigations.

Watermarking and Transparency

Watermarking and transparency are two major components of the South Korean AI Safety Law, formally known as the Act on the Promotion of the Intelligent Information Society and the Management of Artificial Intelligence (AI Basic Act). The legislation focuses on fostering trust and safety by requiring developers and operators of generative AI and high-impact AI to implement specific transparency measures.  

Key Details

• Mandatory Labeling: The law requires that content generated by AI (images, video, or audio) must be clearly labeled to indicate it is AI-generated.

  
  

• Watermarking Requirements: The draft enforcement decree, released in September 2025, specifically permits the use of both human-readable and machine-readable (invisible) watermarks to satisfy these requirements.

  
  

• Deepfake Controls: Stricter rules apply to content that is difficult to distinguish from reality, such as deepfakes, which require conspicuous labeling as a shield against misinformation.

  
  

• User Notice: Operators must provide advance notice to users when they are interacting with generative AI or high-impact AI systems.

  
  

• Exceptions: These obligations may be exempt when using AI for internal business purposes or if the use is already obvious to the user. 

  
  

The law applies to any company that affects the South Korean market, regardless of its headquarters' location.

• Local Representatives: Global giants like OpenAI and Google (which meet specific revenue or user thresholds) are now required by law to designate a representative in Korea to handle safety compliance and government probes.

• AI Safety Research Institute: The law established a new national body to conduct "Red Teaming" (stress-testing) on advanced models to prevent catastrophic risks.

  
  

Unlike the EU's more restrictive approach, Korea’s law intentionally avoids banning specific AI categories.

The law includes a "Master Plan" updated every three years to fund AI startups and infrastructure.

Regulators hold fine thresholds at 30 million KRW. (~$22,000 USD). This is a "guidance-first" approach compared to the EU's multi-million-euro penalties.

While the law is "active," the government has announced a 12-month window focused on education, enabling a transition before strict punitive measures take effect.